Initiativen Für Die Wissenschaft 2018 VORWORT

JAHRES BERICHT Initiativen für die Wissenschaft 2018 VORWORT

Der digitale Wandel ist sprichwörtlich. Er erfasst alle Bereiche unseres Lebens, sei es Wirtschaft, allgemeine Fragen der Gesellschaft, Politik, Kunst und Kultur. Besonders betroffen ist die Wissen- schaft. Dafür gibt es mehrere Gründe. Zum einen sind inzwischen aus den digitalen Techniken eine Reihe von wissenschaftlichen Methoden entstanden, man denke nur an Simulation und Daten- analyse. Zum anderen sind zentrale Fragen der Wissenschaft betroffen, wie die Archivierung und PROF. DR. DR. H.C. Verbreitung von Wissen, genau wie die wissenschaftliche Diskussion, die Form der Kommu- MANFRED BROY nikation zwischen den Wissenschaftlern – alles Themen im Herzen der Wissenschaft. Ein Ergebnis ist, dass in nahezu allen Fächern Methoden der Informatik und der digitalen Technik eine immer lichen Disziplinen und die sich daraus ergebenden stärkere Rolle spielen. Möglichkeiten zur Gewinnung neuer Perspektiven und einer entscheidenden Weiterentwicklung. Dadurch verändern sich Wissenschaften zwar in- haltlich schnell, jedoch ist die Wissenschaftsorga- Dies wird in dem vorliegenden Bericht anschau- nisation nicht immer auf schnelle Veränderungen lich dargestellt. Mein Dank gilt den Mitarbeitern ausgerichtet. Vor diesem Hintergrund hat es das des ZD.B, allen voran Frau Nina Höhne und Herrn Zentrum Digitalisierung.Bayern als eine seiner Daniel Méndez, die nicht nur die wissenschaft- wichtigen Aufgaben begriffen, dem Wissenschafts- lichen Initiativen vorangetrieben, sondern auch leben dabei zu helfen – was die Digitalisierung diesen Band zusammengestellt und ermöglicht betrifft – ein wenig mehr Schwung zu gewinnen. haben. Zu betonen ist hierbei, dass dieser Band Dies wurde erreicht, indem 20 neue Professuren ursprünglich im Rahmen eines Symposiums der an bayerischen Hochschulen und Universitäten Initiativen für die Wissenschaft im Oktober 2018 geschaffen, Nachwuchsforschungsgruppen, ein entstanden ist und nun in einer Form vorliegt, die Doktorandenprogramm, Innovationslabore und einen umfassenden Einblick in die wissenschaft- das Thema der Entrepreneurship-Ausbildung umlichen Initiativen des ZD.B erlaubt. In diesem fassender angegangen wurden. Zusammenhang gilt unser Dank auch der Baye- rischen Staatsregierung und allen voran dem Bay- Das Ergebnis ist eine umfangreiche Initiative, in erischen Staatsministerium für Wissenschaft und der die einzelnen Maßnahmen aufeinander ab- Kunst, die mit ihren entschlossenen Schritten die gestimmt sind und sich gegenseitig befeuern. Der Bildung der Initiativen für die Wissenschaft erst vorliegende Band zeigt dies mit großem Nach- ermöglicht und diese nachdrücklich unterstützt druck. Was geschaffen wurde, ist eine wissen- haben. schaftliche Gemeinschaft in Bayern zum Thema Digitalisierung, die eine ganze Reihe von Fragen in diesem Bereich in Angriff nimmt und dabei bewusst alle Teile der Hochschulen und Wissen- schaften einbezieht: Studierende, wissenschaftliche Mitarbeiter, Promovierende und Professo- rinnen oder Professoren. Besonders wichtig ist Manfred Broy hierbei der Austausch zwischen den unterschied- im November 2018

2 VORWORT | 3 EINLEITUNG Ein solcher Verbund versetzt uns in die Lage gagement aller beteiligten Wissenschaftlerinnen • die bayerische Hochschullandschaft und Wirt- und Wissenschaftler, beispielswiese durch das schaft zu vernetzen, um Bayern nachhaltig zu Engagement in den ZD.B Gremien, möglich. Es Themen der Digitalisierung zu stärken ist eine außerordentliche Freude zu sehen, wie, • Multiplikatoren zu Themen der Digitalisierung dank der Unterstützung verschiedener Partner in die Hochschullandschaft in Hinblick auf For- aus der Wissenschaft, unsere Förderprogramme schung und Lehre zu bilden durch die Gewinnung kluger Köpfe mit Leben ge- • interdisziplinäre Fragestellungen der Digitalisie- füllt werden konnten und wie gut sich die durch rung im Rahmen intensiv kooperierender For- die Forscherinnen und Forscher adressierten scherinnen und Forscher unterschiedlicher Diszi- Themen bereits jetzt ergänzen. plinen und in Zusammenarbeit mit den Experten und Partnern der ZD.B Themenplattformen zu Mit diesem ersten Jahresbericht der ZD.B Initia- PD DR. DANIEL MÉNDEZ DR. NINA HÖHNE durchdringen. tiven für die Wissenschaft möchten wir auf die Ergebnisse der Maßnahmen zurückblicken und Das ZD.B versteht sich dabei als zentrale und unseren Dank an alle Beteiligten für die ange- interdisziplinär ausgerichtete Kollaborations- nehme, inspirierende und wertvolle Zusammen- und Forschungsplattform, die es erlaubt, Wis- arbeit zum Ausdruck bringen. senschaftlerinnen und Wissenschaftler aus dem unmittelbaren Umfeld des ZD.B jenseits diszi- Wir freuen uns darauf, auch 2019 unsere Vision plinärer Grenzen zusammenzuführen und Kom- gemeinsam mit allen beteiligten Akteuren wei- petenzen zu bündeln. Nur mittels eines solchen ter mit Leben zu füllen und die Digitalisierung in Mit der Gründung des Zentrum Digitalisierung. Während die ZD.B Professuren von vorneherein Katalysators kann die Digitalisierungsoffensi- Bayern begeistert voranzutreiben. Bayern (ZD.B) vor drei Jahren stellte sich bei der mit spezifischen fachlichen Widmungen verse- ve des Freistaates langfristig und institutionell Konzeption der Initiativen für die Wissenschaft hen wurden, sind die Programme für die Förde- übergreifend etabliert, verstetigt und nutzbar von Beginn an die strategische Frage: Wie lassen rung des wissenschaftlichen Nachwuchses ganz gemacht werden. sich die anfänglich punktuellen, regional verteil- bewusst thematisch offen geblieben. Die ZD.B ten Fördermaßnahmen für die Wissenschaft, die Professuren wurden dazu im wettbewerblichen Motiviert durch die Erfolge bei der Umsetzung im Rahmen des ZD.B an einzelnen Institutionen Verfahren an die Universitäten und Hochschulen der Förderprogramme galt es daher, im Jahr Daniel Méndez Nina Höhne und Hochschulen angesiedelt sind, durch ge- für angewandte Wissenschaften vergeben, wo 2018 diese einzelnen Initiativen zusammenzu- zielte Vernetzung und Kollaboration zu einem sie im Rahmen autonomer, hochschulinterner führen und die Vision eines interdisziplinären tragfähigen großen Ganzen machen? Entstanden Berufungsverfahren besetzt wurden. Für die Be- Forschungsinkubators in die Tat umzusetzen. ist die Vision eines interdisziplinären Forschungs- setzung der Nachwuchsforschungsgruppen und Zentrale Maßnahmen waren dabei: inkubators als Netzwerk aller wissenschaftlichen für die Auswahl der Promovierenden wurden un- • Roll-Out des ZD.B Graduate Program mit viel- Akteure des ZD.B. abhängige, außerbayerische Expertengremien seitigen Angeboten für die Promovierenden der gebildet, die im Rahmen thematisch offen gehal- wissenschaftlichen Initiativen Die Grundlage dafür sind die durch die Bayerische tener Ausschreibungsverfahren die personenbe- • Zusammenführung der ZD.B Akteure für die Staatsregierung neu geschaffenen Förderpro- zogenen Anträge begutachtet haben. Wissenschaft und Vernetzung durch interne gramme im Rahmen des ZD.B zur Stärkung der Netzwerkveranstaltungen Forschung und Lehre im Bereich der Digitalisie- Übergeordnetes Ziel der Initiativen für die Wissen- • Etablierung der ZD.B Arbeitskreise zur themen- rung. Diese Maßnahmen werden über das ZD.B schaft ist es, die Forschung und Lehre auf allen bezogenen und institutionsübergreifenden in- in Szene gesetzt und zentral koordiniert: Ebenen zu Themen der Digitalisierung zu stärken, haltlichen Arbeit • Die Berufung von 20 neu geschaffenen ZD.B Pro- Multiplikatoreffekte zu erzeugen und den Nähr- • Durchführung des ersten ZD.B Symposiums fessuren (10 an Universitäten und 10 an Hoch- boden für ein generationenübergreifendes Netz- als wesentlicher Meilenstein von den Förder- schulen für angewandte Wissenschaften) werk aus Wissenschaftlerinnen und Wissenschaft- programmen hin zur Etablierung des ZD.B als • Die Etablierung von 10 Nachwuchsforschungs- lern zu schaffen, um Themen der Digitalisierung Forschungsinkubator gruppen wissenschaftlich durchdringen zu können. Jede • Die Etablierung eines Doktorandenprogramms Initiative ist dabei für sich alleine bereits einzig- Unserer Vision einer interdisziplinären Vorge- zur Digitalisierung mit der finanziellen Förderung artig, entfaltet aber gerade im Verbund mit den hensweise zu Digitalisierungsthemen sind wir von 10 Promovierenden sowie einem promoti- anderen Förderprogrammen eine Wirkungskraft, mit diesen Maßnahmen ein gutes Stück näherge- onsbegleitenden und bayernweiten Kurspro- die noch ihresgleichen sucht. kommen – das war vor allem auch durch das En- gramm

4 | EINLEITUNG EINLEITUNG | 5 AKTEURE SEITE 7

ARBEITSKREISE SEITE 40

SYMPOSIUM SEITE 58

6 AKTEURE | 7 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Oliver Amft Prof. Dr.-Ing. Konrad Doll Digital Health Kooperative automatisierte Verkehrssysteme Friedrich-Alexander-Universität Erlangen-Nürnberg Hochschule für angewandte Wissenschaften Aschaffenburg http://www.cdh.med.fau.de https://www.h-ab.de/ueber-die-hochschule/organisation/labor/ kooperative-automatisierte-verkehrssysteme/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Der Lehrstuhl Digital Health forscht an der medizinischen men in Alltagsgegenständen, wie Accessoires und Kleidung. Individuelle Mobilität ist für unsere Gesellschaft von zentraler Verkehrsteilnehmern wie Fußgängern und Radfahrern. Wir Wissensgewinnung und neuen Markern die durch innovative Spezifische Schwerpunkte des Lehrstuhls sind Algorithmen Bedeutung. Sie hat aber Verkehrsunfälle und den Verbrauch erforschen Bewegungsmodelle dieser Verkehrsteilnehmer, mobile und ubiquitäre Systeme zur Verhaltens-, Physiologie- zur Personalisierung von Erkennungs- und Schätzmodellen von Ressourcen zur Folge. Im Forschungsgebiet kooperative um deren Absichten möglichst frühzeitig zu erkennen und und Kontexterkennung gewonnen werden, einschließlich der durch dynamisch gewonnene Informationen, die Integration automatisierte Verkehrssysteme geht es darum, diese Nach- um deren Trajektorien zu prädizieren. Für die Perzeption und Sensorik, klinische Informationsintegration, Musteranalyse- von nicht-überwachten Lernmethoden und Expertenmodel- teile erheblich zu reduzieren. Wir konzentrieren uns auf das Prädiktion verwenden wir u. a. sowohl analytische Methoden und KI-Algorithmen, sowie volldigitalen Entwicklungsprozes- len, sowie Methoden zur Personalisierung der Messtechnik automatisierte Fahren im urbanen Bereich. Eine besonde- als auch Methoden des maschinellen Lernens, kooperative sen zur Herstellung von integrierten elektronischen Syste- (Wearables) mit additiven Fertigungsverfahren. re Herausforderung ist dabei der Umgang mit verletzlichen Vorgehensweisen und Sensordatenfusion.

KURZER LEBENSLAUF KURZER LEBENSLAUF

Oliver Amft, Jahrgang 1975, hat an der TU Chemnitz Elektro- Positionen als Assistant Professor im Bereich Signalverarbei- Konrad Doll, Jahrgang 1964, hat an der Technischen Univer- fenburg berufen. Seit Oktober 2016 hat er die Z.DB Profes- technik studiert und zwischen 2000 und 2004 R&D und Lei- tung an der TU Eindhoven bis 2013 und erster Leiter des W3 sität München studiert und 1994 mit Auszeichnung promo- sur „Kooperative automatisierte Verkehrssysteme“ inne. Er tungspositionen bei ABB begleitet. 2008 promovierte er an Lehrstuhls für Sensorik an der Uni Passau. Seit Okt 2017 leitet viert. Dafür erhielt er den Dissertationspreis der Gesellschaft hat erfolgreich mehrere Forschungsprojekte, die sowohl öf- der ETH Zürich mit Auszeichnung in den Bereichen Wearable er den neu gegründeten ZD.B Lehrstuhl für Digital Health für Informatik in dem Bereich „Anwendungen und Implikatio- fentlich als auch direkt von der Industrie gefördert wurden, Computing, Musteranalyse und Biomedizin. Danach folgten an der Friedrich Alexander Universität Erlangen-Nürnberg. nen der Informatik“. Anschließend war er bei der Motorola durchgeführt. Seine Forschungsarbeiten wurden mehrfach GmbH beschäftigt. Im Jahr 1997 wurde er an die HAW Aschaf- für Best Paper Awards nominiert und ausgezeichnet. AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN 01. Zhang, R., Amft, O.: Monitoring chewing and eating in free-living using smart eyeglasses, IEEE Journal of Biomedical and Health Informatics, 2018 01. Zernetsch, S., Kress, V., Sick, B., Doll, K.: Early Start Intention Detection of Cyclists Using Motion History Images and a 02. Wahl, F., Zhang, R., Freund, M., Amft, O.: Personalizing 3D-printed smart eyeglasses to augment daily life, IEEE Computer, 2017 Deep Residual Network, IEEE Intelligent Vehicles Symposium, 2018 03. Altini, M., Casale, P., Penders, J., Amft, O., et al.: Cardiorespiratory fitness estimation using wearable sensors: Laboratory 02. Goldhammer, M., Köhler, S., Doll, K., Sick, B.: Track-based Forecasting of Pedestrian Behavior by Polynomial Approxima- and free-living analysis of context-specific submaximal heart rates, Journal of Applied Physiology, 2016 tion and Multilayer Perceptrons, IntelliSys 2015, Springer International Publishing, 2016 03. Köhler, S., Goldhammer, M., Bauer, S., Doll, K., et al.: Stationary Detection of the Pedestrian's Intention at Intersections, #Mobile Gesundheit #Ubiquitous Computing #Wearables #Sensorik #Biomedizinische Technik IEEE Intelligent Transportation Systems Magazine, 2013

#Automatisiertes Fahren #verletzliche Verkehrsteilnehmer #Perzeption #Prädiktion #maschinelles Lernen

8 | AKTEURE STECKBRIEFE | 9 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Andreas Festag Prof. Dr. Jens Grubert Fahrzeugsicherheit und Car2X-Kommunikation Mensch-Maschine-Interaktion im Technische Hochschule Ingolstadt Internet der Dinge http://bit.ly/Andreas_Festag Hochschule für angewandte Wissenschaften Coburg https://www.mixedrealitylab.de

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Der Einsatz von Informations- und Kommunikationstechno- Verkehrsinfrastruktur zur Erhöhung der Verkehrssicherheit Die stetig steigende Anzahl an smarten Geräten sowie überall (Virtual Reality, VR) um mittels diesen mit teils physisch – logien verbessert die Leistungsfähigkeit und die Effizienz der genutzt werden kann. Kernthemen umfassen das Design, und ständig verfügbaren digitalen Informationen verlangen teils digitalen Produkten und Dienstleistungen interagieren Verkehrssysteme nachhaltig. Die hochgradige Vernetzung die Leistungsbewertung und die Optimierung der Kommu- nach neuen Mitteln diese für Menschen zugänglich und nutz- zu können. Beispiele umfassen den Einsatz der AR für die in Intelligenten Verkehrssystemen ist die Basis für eine Viel- nikationssysteme für Car2X Use Cases. Herausforderungen bar zu machen. Die Gruppe von Prof. Dr. Jens Grubert er- Fernwartung im Maschinenbau, das effiziente Training von zahl von Anwendungen. Das Forschungsgebiet Fahrzeug- sind der Einsatz von Car2X zur Unterstützung der Fahrzeug- forscht Grundlagen und Anwendungen der vermischten Rea- Handlungsabläufen mittels AR/VR in Industrieanlagen oder sicherheit und Car2X Kommunikation untersucht, wie der automatisierung sowie die Integration aktiver und passiver lität (Mixed Reality) mit seinen Ausprägungen der erweiterten die Erforschung des virtuellen Büros der Zukunft. Informationsaustausch zwischen Fahrzeugen sowie mit der Sicherheitsfunktionen im Fahrzeug. Realität (Augmented Reality, AR) und der virtuellen Realität

KURZER LEBENSLAUF KURZER LEBENSLAUF

Andreas Festag, Jahrgang 1968, studierte Elektrotechnik und Infrastruktursysteme. Prof. Dr. Festag ist seit vielen Jahren in Jens Grubert, Jahrgang 1983, ist seit 2016 Professor für Mensch- sein Diplom der Computervisualistik mit Auszeichnung an promovierte im Bereich Telekommunikationsnetze, beides der Standardisierung für Car2X aktiv, u.a. im ETSI Technical Maschine-Interaktion im Internet der Dinge an der Hoch- der Otto-von-Guericke Universität Magdeburg. Seine akade- an der Technischen Universität Berlin. Danach arbeitete er Committe ITS, im Car-2-Car Communication Consortium und schule Coburg. Zuvor arbeitete als Post-Doc an der Uni- mischen und industriellen Erfahrungen resultierten in über in verschiedenen Positionen am Heinrich-Hertz-Institut in in der 5G Automotive Association. Seit März 2018 ist er Inha- versität Passau. Er promovierte 2015 mit Auszeichnung bei 50 begutachteten Publikationen und Patenten, darunter ein Berlin, NEC Laboratories in Heidelberg, Technischen Uni- ber der ZD.B-Professur “Fahrzeugsicherheit und Car2X-Kom- Prof. Dr. Dieter Schmalstieg im Bereich Augmented Reality Praxisbuch zum Thema AR Entwicklung für Android. versität Dresden und Fraunhofer Institut für Verkehrs- und munikation” an der Technischen Hochschule Ingolstadt. an der Technischen Universität Graz, Österreich und erhielt

AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN

01. Zhang D., Festag A., Fettweis G.: Performance of Generalized Frequency Division Multiplexing Based Physical Layer in 01. Grubert, J., Langlotz, T., Zollmann, S., Regenbrecht, H.: Towards pervasive augmented reality: Context-awareness in Vehicular Communication, IEEE Transactions on Vehicular Technology, 2017 augmented reality, IEEE transactions on visualization and computer graphics, 23(6), 2017 02. Festag A.: Standards for Vehicular Communication – From IEEE 802.11p to 5G, e & i Elektrotechnik und Informationstechnik, 02. Grubert, J., Kranz, M.: Headphones: Ad hoc mobile multi-display environments through head tracking, CHI Conference 2015 on Human Factors in Computing Systems, 2017 03. Festag A., Papadimitratos P., Tielert T.: Design and Performance of Secure Geocast for Vehicular Communication, 03. Grubert, J., Heinisch, M., Quigley, A., Schmalstieg, D.: Multifi: Multi fidelity interaction with displays on and around the IEEE Transactions on Vehicular Technology, 2010 body, 33rd Annual ACM Conference on Human Factors in Computing Systems, 2015

#Fahrzeugsicherheit #Vernetzung #Car2X #ITS-G5 #Cellular V2X #Augmented Reality #Virtual Reality #Computer Vision #Human-Computer-Interaction #Smart Glasses

10 | AKTEURE AKTEURE | 11 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Sascha Hauke Prof. Dr. Dominik Herrmann Intelligente Energienetze Privatsphäre und Sicherheit Hochschule für angewandte Wissenschaften Landshut in Informationssystemen http://bit.ly/Sascha_Hauke Otto-Friedrich Universität Bamberg https://www.uni-bamberg.de/psi/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Die Energiewende stellt uns vor zahlreiche Herausforderun- des maschinellen Lernens zur Vorhersage von Komponen- Informationssysteme durchdringen unseren Alltag. Eine an- und Folgen abzuschätzen – es mangelt an Transparenz. Wir er- gen. Um diesen zu begegnen, müssen unsere Stromnetze tenlebensdauern, wie zum Beispiel der Restlebenszeit von gemessene Absicherung der verarbeiteten Daten und der forschen daher, wie sich Informationssystemen mit gut benutz- intelligenter werden. Im Forschungsgebiet Intelligente Ener- Batterien, oder Produktions- und Verbrauchsprognose in Systeme ist entscheidend. Dennoch werden kontinuierlich Si- baren technischen Mechanismen absichern lassen. Darüber gienetze betrachten wir deshalb, wie Stromerzeugung, -spei- Netzen mit verteilter Stromerzeugung. Eine besondere Her- cherheitslücken entdeckt, die entweder aus Unkenntnis und hinaus analysieren wir existierende Systeme und stellen die cherung und -verteilung durch Methoden der Informatik und ausforderung liegt dabei darin, dass bestehende, gewachse- Unfähigkeit oder aber durch Nachlässigkeit entstehen. Es feh- Ergebnisse leicht nachvollziehbar dar (z.B. auf PrivacyScore.org). der IT effizienter, zuverlässiger und sicherer und gestaltet ne Infrastrukturen berücksichtigt werden müssen und diese len wirksame Anreize, Sicherheit und Datenschutz ernst zu Dabei kooperieren wir mit Arbeitsgruppen aus dem Bereich des werden können. Hierzu nutzen wir zum Beispiel Verfahren nur in stark begrenztem Umfang verändert werden können. nehmen. Zudem fällt es Nutzern immer schwerer, Sicherheit maschinellen Lernens, Rechtswissenschaftlern, und Ethikern.

KURZER LEBENSLAUF KURZER LEBENSLAUF

Sascha Hauke, Jahrgang 1977, hat an der Westfälischen Wil- (Abschluss 2015 mit Auszeichnung). In den Jahren 2016 Dominik Herrmann, Jahrgang 1983, hat an der Universität formatik). Danach vertrat er eine Professur für IT-Sicher- helms-Universität Münster Informatik studiert. Nach einem und 2017 war er Mitgesellschafter der compredict GmbH, Regensburg und am University College Dublin Wirtschafts- heitsmanagement an der Universität Siegen (2015–2017). Forschungsaufenthalt an der University of New Brunswick, die 2016 den Hauptpreis und den Sonderpreis "Big Data" informatik studiert. Ab 2008 arbeitete er als Wissenschaft- Seit Oktober 2017 baut er den ZD.B-Informatiklehrstuhl für Kanada, und Forschungs- und Lehrtätigkeiten an der Fach- des Gründerwettbewerbs “Digitale Innovation” des BmWI licher Mitarbeiter in Regensburg, später an der Universität Privatsphäre und Sicherheit in Informationssystemen an der hochschule der Wirtschaft in Bergisch-Gladbach, hat er ab gewann. Seit April 2017 ist er Inhaber der ZD.B-Professur Hamburg. Herrmann hat über die Privatsphäre im Domain Universität Bamberg auf. Herrmann engagiert sich als einer 2011 an der Technischen Universität Darmstadt, Fachbe- “Intelligente Energienetze” an der HAW Landshut. Name System promoviert (Abschluss 2014, ausgezeichnet der Junior Fellows in der GI, u.a. als Herausgeber des News- reich Informatik, bei Prof. Dr. Max Mühlhäuser promoviert u.a. mit dem GI-Dissertationspreis der Gesellschaft für In- letters „GI-Radar“.

AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN

01. Grube T., Hauke S., Daubert J., Mühlhäuser M.: Ant Colony Optimisation - A Solution to Efficient Anonymous Group 01. Maass, M., Wichmann, P., Herrmann,D.: PrivacyScore: Improving Privacy and Security via Crowd-Sourced Benchmarks of Communication?, 14th Annual IEEE Consumer Communications & Networking Conference (CCNC), 2017 Websites, ENISA Annual Privacy Forum, 2017 02. Garcia Cordero C., Hauke S., Mühlhäuser M., Fischer M.: Analyzing Flow-based Anomaly Intrusion Detection using 02. Herrmann, D., Lindemann, J.: Obtaining personal data and asking for erasure: Do app vendors and website owners ho- Replicator Neural Networks, 14th Annual Conference on Privacy, Security and Trust (PST), 2016 nour your privacy rights?, GI SICHERHEIT 2016: Sicherheit – Schutz und Zuverlässigkeit, 2016 03. Hauke S., Biedermann S., Mühlhäuser M., Heider D.: On the Application of Supervised Machine Learning to 03. Kirchler, M., Herrmann, D., Lindemann, J., Kloft, M.: Tracked Without a Trace: Linking Sessions of Users by Unsupervised Trustworthiness Assessment, 12th IEEE International Conference onTrust, Security and Privacy in Computing and Learning of Patterns in Their DNS Traffic, 9th ACM Workshop on Artificial Intelligence and Security (AISec), co-located Communications (IEEE TrustCom-13), 2013 with the 23rd ACM Conference on Computer and Communications (CCS), 2016

#Smart Grid #Security & Safety #Strom #Netze #Privacy Enhancing Technologies #Traffic Analysis #Web Security #Usable Privacy

12 | AKTEURE AKTEURE | 13 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Martin Matzner Prof. Dr. Jürgen Mottok Digital Industrial Service Systems Safe and Secure Systems of Systems Friedrich-Alexander-Universität Erlangen-Nürnberg Ostbayerische Technische Hochschule Regensburg https://www.is.rw.fau.de www.las3.de

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Wir entwerfen Methoden und Werkzeuge für die Entwick- managements, Service Engineering und Service Management Das Laboratory for Safe and Secure Systems (LaS³) ist ein For- stützt. Das LaS³ sieht sich als Mediator zwischen Wissenschaft lung und den Betrieb informationstechnikbasierter Dienst- sowie Informationsmodellierung. Die in unserer Forschung schungscluster der Ostbayerischen Technischen Hochschule und Anwendung. Das LaS³ hat derzeit 15 Doktoranden und leistungssysteme, insbesondere in der Industrie. Dazu nut- erzielten Ergebnisse sollen unmittelbare Beiträge zur Lösung Regensburg. Es wurde 2005 an der damaligen Fachhochschule 10 Studierende des Masterstudiengangs „Master of Applied zen wir Techniken des Geschäftsprozessmanagements und konkreter betrieblicher Probleme und gesellschaftlicher He- Regensburg gegründet und wurde initial durch die Projekt- Research“. In der Vergangenheit wurden bereits etliche For- der Informationsmodellierung. Unsere Forschungsgebiete rausforderungen leisten. förderung FHprofUnt der Bundesrepublik Deutschland sowie schungsprojekte aus öffentlichen als auch aus privaten Dritt- umfassen analytische Anwendungen des Geschäftsprozess- durch das Programm Strukturimpuls des Bayerischen Staats- mitteln erfolgreich durchgeführt. Das LaS³ der OTH Regens- ministeriums für Wissenschaft, Forschung und Kunst unter- burg steht als Innovator für „Safe and Secure Systems”. KURZER LEBENSLAUF KURZER LEBENSLAUF Martin Matzner, Jahrgang 1982, ist Professor für Wirtschafts- an Forschungsprojekten durchgeführt, die durch EU, Bund, informatik und Inhaber des Lehrstuhls für Digital Industrial Länder und Unternehmen finanziert wurden. Er hat mehr als ZD.B-Forschungsprofessor Dr. Jürgen Mottok lehrt seit 2004 der Bayerischen Hochschulen, zeichnet verantwortlich als Service Systems an der FAU Erlangen-Nürnberg. 2012 wurde 80 Forschungsartikel und Konferenzartikel veröffentlich, die Informatik an der OTH Regensburg. Seine Lehrgebiete sind Vorsitzender des Lehren von Software Engineering e.V. (LeSE er für seine Arbeit zum Geschäftsprozessmanagement in unter anderem in MIS Quarterly und IEEE Transactions on Software Engineering, Programmiersprachen, Echtzeitsyste- e.V.) und als Projektleiter mit zahlreichen kooperativen Pro- Service-Netzwerken an der WWU Münster promoviert. 2016 Engineering Management erschienen sind. Er ist Mitheraus- me, Functional Safety und IT-Security. Er leitet wissenschaft- motionsverfahren. Jürgen Mottok ist in Programmkomitees erhielt er die Habilitation für Forschungsarbeiten zu Predic- geber des Journal of Service Management Research. lich das Laboratory for Safe and Secure Systems (LaS³) in zahlreicher wissenschaftlicher Konferenzen vertreten. Er ist tive-Process-Mining. In diesen Bereichen hat er eine Vielzahl Regensburg, ist Vorstandsmitglied des Bavarian Cluster of IT Träger des Preises für herausragende Lehre, der vom Baye- Security and Safety, firmiert als Beirat des Automotive Forum rischen Staatsministerium für Wissenschaft, Forschung und Sicherheit Software Systeme sowie als Beirat des ASQF Sa- Kunst im Jahr 2010 vergeben wurde. Am 04.12.2015 wurde AUSGEWÄHLTE PUBLIKATIONEN fety, ist Mitglied des Leitungsgremiums der Regionalgruppe Prof. Dr. Jürgen Mottok der „Preis für besondere Leistungen Ostbayern der Gesellschaft für Informatik, wirkt als Organi- bei der Zusammenarbeit zwischen Wirtschaft und Wissen- 01. Breuker, D., Matzner, M., Delfmann, P., Becker, J.: Comprehensible Predictive Models for Business Processes, sator des Fachdidaktik-Arbeitskreises Software Engineering schaft“ verliehen. MIS Quarterly, 2016 02. Beverungen, D., Müller, O., Matzner, M., Mendling, J., vom Brocke, J.: Conceptualizing Smart Service Systems, Electronic Markets, 2017 AUSGEWÄHLTE PUBLIKATIONEN 03. Matzner, M., Büttgen, M., Demirkan, H., Neely, A. et al.: Digital Transformation in Service Management, Journal of Service Management Research, 2018 01. Osinski, L., Langer, T., Mottok, J.: A Survey of Fault Tolerance Approaches at Different Architecture Levels, 30th Interna- tional Conference on Architecture of Computing Systems, 2017 #Smart Service #Service Engineering #Business Process Management (BPM) #Process Mining 02. Krämer, S., Raab, P., Mottok, J., Racek, S.: Comparison of Enhanced Markov Models and Discrete Event Simulation, 17th #Robotic Process Automation (RPA) Euromicro Conference on Digital System Design (DSD), 2014 03. Sailer, A., Deubzer, M., Lüttgen, G., Mottok, J.: CoreTAna: A Trace Analyzer for Reverse Engineering Real-Time Software, IEEE 23rd International Conference on Software Analysis, Evolution, and Reengineering (SANER), 2016

#Software Engineering #Functional Safety #IT-Security #Real-Time Systems #Software Engineering Education

14 | AKTEURE AKTEURE | 15 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Jörg Müller Prof. Dr. Nicholas H. Müller Serious Games Sozioinformatik und gesellschaftliche Universität Bayreuth Aspekte der Digitalisierung http://www.ai8.uni-bayreuth.de/de/index.html Hochschule für angewandte Wissenschaften Würzburg-Schweinfurt https://fiw.fhws.de/fakultaet-iw/personen/details/person/ prof-dr-nicholas-mueller.html

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Der Lehrstuhl für Serious Games beschäftigt sich haupt- mithilfe von dynamischen Systemmodellen, sowie biome- Die Professur Sozioinformatik beschäftigt sich anhand sowohl quantitativ als auch qualitativ analysiert. Darüber hin- sächlich mit der Interaktion zwischen Mensch und Com- chanische Simulation der Mensch-Computer Interaktion, empirischer Methoden mit den Herausforderungen der aus werden an der Professur Untersuchungen zu neuen Einga- putern, mit Computerspielen und Serious Games. Unsere Ultraschall-Levitationsschnittstellen und erweiterte sowie Mensch-Computer-Interaktion sowie der Digitalisierung. Dazu bekanälen, wie der Pupillometrie, zur Messung der kognitiven primären Forschungsgebiete sind die Modellierung, Simu- virtuellen Realität. werden u.a. Untersuchungen mittels der Erfassung physiolo- Belastung durchgeführt, virtuelle Realitätserfahrungen durch lation und Optimierung der Mensch-Computer Interaktion gischer Parameter wie der Hautleitfähigkeit, Herzratenvaria- die Implementierung von Eye-Tracking optimiert und neue An- bilität oder Augenbewegungen durchgeführt und die empiri- sätze zur digitalen Unterstützung von Maschinenreparaturen KURZER LEBENSLAUF schen Fragestellungen über eine methodische Triangulation mittels holografischer Darstellungsgeräte entwickelt.

Jörg Müller, Jahrgang 1980, hat an der Albert-Ludwigs-Uni- fessor am Alexander von Humboldt Institut für Internet und KURZER LEBENSLAUF versität Freiburg und der Universität des Saarlandes Infor- Gesellschaft in Berlin. 2014 wurde er Associate Professor matik studiert und an der Universität Münster promoviert. an der Universität Aarhus in Dänemark. Seit August 2017 Nicholas Müller promovierte 2010 an der Professur E-Lear- der Deutschen Flugsicherung an einem Modell zur Berech- Danach arbeitete er als Senior Researcher an den Deutsche ist er Inhaber der ZD.B-Professur „Serious Games“ an der ning der Technischen Universität Chemnitz. Anschließend nung kognitiver Arbeitsbelastungen von Fluglotsen. Ferner Telekom Laboratories in Berlin, sowie ab 2012 als Gastpro- Universität Bayreuth. arbeitete er bei Accenture als Senior Learning Consultant unterstützte er das DFG-Graduiertenkolleg „CrossWorlds“ und wechselte 2012 an die TU Chemnitz, wo er 2016 an der und initiierte die Entwicklung von nonverbalen Mensch-Com- Professur Medienpsychologie habilitierte. In verschiedenen puter-Inputkanälen im Projekt „Human-Responsive-Design“. AUSGEWÄHLTE PUBLIKATIONEN interdisziplinären Projekten arbeitete er u.a. an der Simula- Darüber hinaus konnte er als Start-Up-Mentor bereits zwei tion von Emotionsvalenzen sowie in Zusammenarbeit mit EXIST-geförderte Gründungen realisieren. 01. Müller, J., Oulasvirta, A., Murray-Smith, R.: Control Theoretic Models of Pointing, ACM Transactions on Computer-Human Interaction, 2017 02. Müller, J., Geier, M., Dicke, C, Spors, S.: The BoomRoom: Mid-air Direct Interaction with Virtual Sound Sources, AUSGEWÄHLTE PUBLIKATIONEN SIGCHI Conference on Human Factors in Computing Systems, 2014 03. Fender, A., Herholz, P., Alexa, M., Müller, J.: OptiSpace: Automated Placement of Interactive 3D Projection Mapping 01. Müller, N. H., Liebold, B., Pietschmann, D., Ohler, P., Rosenthal, P.: Visualizations for Hierarchical Data: Content, CHI Conference on Human Factors in Computing Systems, 2018 Analyzing User Behavior and Performance with Eye Tracking, International Journal on advances in Software, 2017 02. Rosenthal, P., Pfeiffer, L., Müller, N.H., Valtin, G.: The Long Way to Intuitive Visual Analysis of Air Traffic Control Data, #Serious Games #Mensch-Computer-Interaktion #Ultraschall-Levitation #Erweiterte und Virtuelle Realität Chapter 11, In: Visual Analytics for Management: Translational Science and Applications in Practice, Routledge, 2016 03. Pfeiffer, L., Valtin, G., Müller, N. H., Rosenthal, P.: The Mental Organization of Air Traffic and its Implications to an -Emo tion Sensitive Assistance System, International Journal On Advances in Life Sciences, 2016

#Sozioinformatik #Digitalisierung #Medienwirkung #Physiologische Messungen #Mensch-Computer-Interaktion

16 | AKTEURE AKTEURE | 17 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Andreas Muxel Prof. Dr.-Ing. Eva Rothgang Physical Human-Machine Interfaces Digitale Prozessketten in der medizinischen Hochschule für angewandte Wissenschaften Augsburg Versorgung und Medizintechnik https://hybridthings.hs-augsburg.de Ostbayerische Technische Hochschule Amberg-Weiden https://www.oth-aw.de/rothgang

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

War die Rollenverteilung in der Vergangenheit noch klar – aktuell kontrovers diskutiert. Es gilt hierbei das Vertrauen Digitalisierung bietet enorme Potentiale für ein effizientes Erkenntnissen/ Prozessschritten. Um dabei nutzerzentrierte der Mensch handelt, die Maschine funktioniert – erleben wir der Nutzer in neue Technologien zu stärken und sinnvolle Gesundheitssystem und eine individuelle Patientenversor- Entwicklung optimal umsetzen zu können, wird aktuell ein immer mehr, dass wir mit den Dingen »mit-funktionieren« Kommunikations- und Interaktionsformen zwischen Mensch gung. Ein Schwerpunkt ist hier die Vernetzung von Daten User Research & Usability Labor mit Fokus auf Medizin- und diese mit uns »mit-handeln«. So werden wir zukünftig und abstrakter Maschine zu etablieren. Im HYBRID THINGS und daraus resultierend die Generierung von neuartigen technik an der OTH-AW aufgebaut. unseren Arbeitsraum, Verkehrsraum und Lebensraum mit LAB werden zukünftige Szenarien schon heute erlebbar ge- antizipativen und autonomen Systemen teilen. Entsprechen- macht, um potenzielle Anwendungsgebiete im jeweiligen de technische und rechtliche Rahmenbedingungen stellen Kontext zu bewerten und mögliche Chancen, aber auch Ri- KURZER LEBENSLAUF immer noch eine große Herausforderung dar und werden siken aufzuzeigen. Eva Rothgang, Jahrgang 1982, hat an der Universität Ulm und ten, eingebettet ist. Seit Anfang 2012 hat Eva Rothgang in University of Essex, Großbritannien, Informatik mit Schwer- verschiedenen Stationen bei Siemens in der Medizintechnik KURZER LEBENSLAUF punkt Medizin studiert. Ihre Promotion (Doktorvater Prof. gearbeitet und zuletzt das Kompetenzzentrum zum Thema Hornegger, Lehrstuhl für Mustererkennung, FAU) führte sie Workflow im Geschäftsgebiet MR geleitet. Seit Mai 2017 ist Andreas Muxel, Jahrgang 1979 studierte Kommunikationsde- Entwicklung und Gestaltung hybrider Artefakte und Räume. anschließend an das Center for Applied Medical Imaging sie Inhaberin der ZD.B Professur „Digitale Prozessketten in sign an der Fachhochschule Vorarlberg und Mediale Künste an Von 2013 bis 2017 war Andreas Muxel Professor für Inter- (Siemens Corporate Research), das in die Radiologie des der medizinischen Versorgung und Medizintechnik“ an der der Kunsthochschule für Medien Köln. Seine Projekte an der face Design an der Köln International School of Design der TH Johns Hopkins Hospital in Baltimore (MD), Vereinigte Staa- OTH Amberg-Weiden. Schnittstelle von Design, Kunst und Wissenschaft wurden be- Köln. Seit September 2017 ist er Inhaber der ZD.B-Professur reits vielfach international ausgestellt und ausgezeichnet (u.a. “Physical Human-Machine Interfaces” an der Fakultät für Ge- Ars Electronica Festival Linz, FILE Festival São Paulo, TodaysArt staltung der Hochschule Augsburg und leitet dort das HYBRID AUSGEWÄHLTE PUBLIKATIONEN Festival Brüssel, VIDA Award Madrid, Share Price Turin). 2015 THINGS LAB. 01. Denck, J., Landschütz, W., Rothgang, E., et al.: Automated Billing Code Prediction from MRI Log Data, Joint Annual Gründung des Design Studios NEOANALOG mit Schwerpunkt Meeting ISMRM-ESMRMB, 2018 02. Overduin, C.G., Heidkamp, J., Rothgang, E., et al.: Fast 3-T MR-guided transrectal prostate biopsy using an in-room tablet AUSGEWÄHLTE PUBLIKATIONEN device for needle guide alignment: a feasibility study, European Radiology, 2018 03. Rothgang, E., Gilson, W.D., Wacker, F., et al.: Rapid freehand MR-guided percutaneous needle interventions: an ima- 01. Muxel, A.: Der Aufstand der Dinge, S. 51ff, In: Beseelte Dinge – Design aus Perspektive des Animismus, transcript Verlag, ge-based approach to improve workflow and feasibility, Journal of Magnetic Resonance Imaging, 37(5), 2013 Bielefeld 2016 02. Muxel, A., Nawrath, M., Scherffig, L., Trogemann, G.: Experimente, S. 29ff, In: Code und Material – Exkursionen ins #Medizintechnik #Gesundheitswirtschaft #Mensch-Maschine-Interaktion #Mustererkennung #Künstliche Intelligenz Undenkliche, Springer-Verlag, Wien 2010 03. Strauss, W., Fleischmann, M., Muxel, M., Denzinger, J., Himmel, A.: Matrix-Lupe. Design-Präsentation: Navigation und Visualisierung, Proceedings Mensch und Computer, 2006

#Mensch-Maschine-Schnittstellen #Mensch-Roboter-Kollaboration #Internet der Dinge #Tangible Interaction #Affective Computing

18 | AKTEURE AKTEURE | 19 AKTEURE AKTEURE

ZD.B Professur ZD.B Professur

Prof. Dr. Albrecht Schmidt Univ.-Prof. mult. Dr. habil. Human-Centered Ubiquitous Media Björn W. Schuller Ludwig-Maximilians-Universität München Embedded Intelligence for https://www.um.informatik.uni-muenchen.de Health Care and Wellbeing Universität Augsburg, Imperial College London u.A. https://www.informatik.uni-augsburg.de/lehrstuehle/eihw/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Durch die Digitalisierung vollzieht sich im Moment ein Umbruch Benutzung ist die zentrale Herausforderung, welche durch Hauptforschungsfeld ist die sensor- und wissensbasierte Be- sicht, Gestik, Körpermotorik) in einem realistischen Szenario in nahezu allen Bereichen unseres Lebens. Menschen nutzen autonome Systeme noch vergrößert wird. Die übergeordnete gleitung und Überprüfung gesundheitsrelevanter Parameter (Alltagsleben) verarbeitet. Ziel dabei sind Applikationen, die digitale Technologien im Privaten wie im Beruflichen. Kommu- Forschungsfrage lautet: Wie können wir menschliche Fähig- bei sportlichen und anderen Aktivitäten. Das Hauptinteresse entsprechende Parameter erheben und sinnvolle Rückmel- nikation, Wissenserwerb, Dokumentation, Nachrichten, Musik, keiten durch digitale Technologien erweitern? Unser Fokus liegt liegt auf der Erfassung, Analyse und Interpretation von Bio- dungen an den Benutzer geben. Dazu werden neue Algo- Fotografie oder Reiseplanung basieren selbstverständlich auf auf der Bearbeitung von kritischen Fragen in der Grundlagen- signalen, wie sie etwa bei der Überwachung der Herzaktivität, rithmen zur robusten maschinellen Signalverarbeitung und digitaler Technologie. Die Schnittstelle zwischen Mensch und forschung. Im Kontext von Domänen und Anwendungen be- des Stoffwechsels oder der neuronalen Aktivitäten anfallen. zum maschinellen Lernen entwickelt – insbesondere für die digitalen Medien ist allgegenwärtig, und die Einfachheit ihrer trachten wir auch resultierende gesellschaftliche Auswirkungen. Daneben werden auch akustische Parameter (Sprache und vielversprechenden tiefen neuronalen Netzwerke. andere akustische Ereignisse) sowie visuelle Parameter (Ge- KURZER LEBENSLAUF KURZER LEBENSLAUF Albrecht Schmidt hat in Ulm und Manchester Informatik stu- wicklung von Computertechnologie, Software und digitalen diert und hat 2003 an der Lancaster University in Großbri- Medien. Für seine wesentlichen wissenschaftlichen Beiträge Björn W. Schuller, Jahrgang 1975, schloss Diplom (1999), Pro- inhaber an der Universität Passau, Wissenschaftler am CNRS tannien promoviert. Zu seinen beruflichen Stationen zählen in der Mensch-Computer-Interaktion wurde er 2018 in die motion (2006) und Habilitation (2012) in EI an der TUM ab. Er bei Paris, der Joanneum Research in Graz und Gastprofessu- das KIT, das Fraunhofer Institut für Intelligente Informations- ACM SIGCHI Academy aufgenommen. Im Jahr 2016 erhielt er ist Ordinarius an der Universität Augsburg, Professor für K.I. ren in Genf und Vilnius. Er ist Koautor von 750+ Publikationen und Analysesysteme, die Universität Duisburg-Essen, die Uni- vom Europäischen Forschungsrat (ERC) eine fünfjährige För- am Imperial College London, mitbegründender Geschäfts- (20k Zitierungen, h-Index 66), IEEE Fellow, Senior Member versität Stuttgart, die Universität Cambridge und Microsoft derung, um digitale Technologien zur Verstärkung mensch- führer der audEERING GmbH, permanenter Gastprofessor ACM, emeritierter AAAC Präsident sowie Hauptherausgeber Research. Er beschäftigt sich mit der menschzentrierten Ent- licher Kognition und Perzeption zu erforschen. am HIT/China und Berater von Unternehmen wie GN, HUA- des IEEE Journals seines Fachgebiets. WEI und Samsung. Frühere Stationen beinhalten Lehrstuhl- AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN 01. Schmidt, A.: Technologies to Amplify the Mind, IEEE Computer, 50(10), 2017 02. Schmidt, A., Herrmann, T.: Intervention user interfaces: a new interaction paradigm for automated systems, 01. Rudovic, O., Lee, J., Schuller, B.W., et al.: Personalized machine learning for robot perception of affect and engagement in ACM interactions 24, 2017 autism therapy, Science Robotics, AAAS, 2018 03. Karolus, J., Wozniak, P.W., Chuang, L.L., Schmidt, A.: Robust Gaze Features for Enabling Language Proficiency Awareness, 02. Mencattini, A., Mosciano, F., Schuller, B.W., et al.: An emotional modulation model as signature for the identification of CHI Conference on Human Factors in Computing Systems, 2017 children developmental disorders, Nature Scientific Reports, Nature Publishing Group, 2018 03. Trigeorgis, G., Nicolaou, M.A., Zafeiriou, S., Schuller, B.W.: Deep Canonical Time Warping for simultaneous alignment and #Mensch-Technik-Interaktion #Intelligente Benutzungsschnittstellen #Virtuelle und erweiterte Realität representation learning of sequences, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018 #Multimodale Interaktion #Interaktive Medien #Mobile Gesundheit #Medizininformatik #Künstliche Intelligenz #Tiefes Lernen #Signalverarbeitung

20 | AKTEURE AKTEURE | 21 AKTEURE AKTEURE

ZD.B Nachwuchs- ZD.B Professur forschungsgruppe

Prof. Dr. Ramin Tavakoli Kolagari Prof. Dr. Florian Alt Softwareentwicklung für sichere Usable Security and Privacy und autonome Fahrzeugsysteme Universität der Bundeswehr München www.florian-alt.org Technische Hochschule Nürnberg https://www.th-nuernberg.de/fakultaeten/in/forschung/ automotive-software-systems-engineering/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Unter modernen Fahrzeugsystemen verstehen wir intelligente, insbesondere mit Ingenieuren, Juristen und Technikethikern. Computer sind mittlerweile ein fester Bestandteil unseres täg- wir uns mit der Rolle des Menschen in sicherheitskritischen zuverlässige, sichere und vernetzte Softwaresysteme im Auto- Zudem kooperieren wir international mit Hochschulen und lichen Lebens – sowohl in Form persönlicher Geräte als auch Umgebungen. Insbesondere entwickeln wir neuartige Konzepte mobil. Ausgehend von ingenieurstechnischen Fachkenntnis- Partnern aus der Automobil- und Softwareindustrie. Unsere eingebettet in unserer Umgebung. Dies ermöglicht es uns, und Systeme, welche die stetige Veränderung in der Art wie wir sen im Bereich Softwareentwicklung Automotive gelingt es uns, Studierenden erhalten eine umfassende Ausbildung, die sie be- jederzeit und überall mit sensitiven Informationen zu inter- mit Computern interagieren, berücksichtigen. Schwerpunkte neue Konzepte und Methoden zu generieren. Dadurch unter- fähigt, Lösungen für komplexe Problemstellungen in Forschung agieren. Zeitgleich stellt uns der sichere Umgang mit solchen sind verhaltensbiometrische Verfahren, sichere Schnittstellen stützen wir die Umsetzung nachweislich zuverlässiger Software- und Praxis prozessorientiert zu entwickeln. Im Rahmen des sensitiven Informationen vor große Herausforderungen beim in Smart Homes, neuartige Authentifizierungsmechanismen systeme im Automobilkontext. Unsere Arbeitsweise orientiert Service Learning sind wir außerdem bemüht, die Studierenden Design sicherer Systeme. In unserer Forschung beschäftigen sowie Benutzerschnittstellen für IT-Sicherheitsexperten. sich am Leitgedanken der interdisziplinären Zusammenarbeit, zu gesellschaftlichem Engagement zu bewegen. KURZER LEBENSLAUF KURZER LEBENSLAUF Florian Alt ist seit 1. Mai 2018 Professor am Forschungsin- halb Jahre in den USA als Webentwickler und IT-Spezialist Prof. Dr. Tavakoli Kolagari ist seit dem Wintersemester 2011/ nome Fahrzeugsysteme. Das Gebiet umfasst den gesamten stitut für Cyber-Sicherheit der Universität der Bundeswehr gearbeitet. Florian Alt ist General Chair der Konferenz Mobile 2012 an der Technischen Hochschule Nürnberg in der Fakul- Bereich von den frühen Phasen der Softwareentwicklung bis in München mit dem Schwerpunkt „Usable Security and und Ubiquitous Multimedia (MUM 2018), Vorsitzender des tät Informatik tätig und vertrat dort zunächst das Lehrgebiet zur Softwarewiederverwendung und Implementierung. Spe- Privacy“. Davor war er als Juniorprofessor Leiter der ZD.B Programm-Komitees der Konferenz `Mensch und Computer Software Engineering. Seit dem Jahr 2017 vertritt er im Rah- zielles Augenmerk wird hierbei auf die Qualität im Sinne von Nachwuchsgruppe „Biometrics++“ an der HAW München und 2019‘, Associate Chair der ACM SIGCHI Konferenz und Asso- men einer Professur mit enger Anknüpfung an das Zentrum Safety und Security und auf die Anwendungsfunktionalität der LMU München. Florian Alt hat Medieninformatik studiert ciate Editor der Zeitschrift ACM IMWUT und des Magazins Digitalisierung.Bayern (Garching) das Lehr- und Forschungs- autonom fahrender Autos gelegt. Auch mehrere Promotionen und an der Universität Stuttgart promoviert. Er hat einein- IEEE Pervasive Computing. gebiet Softwareentwicklung für sichere und zuverlässige auto- zum Thema Automotive Security werden betreut. AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN 01. Alt, F. Müller, J., Schmidt, A.: Advertising on Public Display Networks, IEEE Computer, 2012 01. van Wagensveld, R., Wägemann, T., Hehenkamp, N., Tavakoli Kolagari, et al.: Intra-Task Parallelism in Automotive 02. Davies, N., Clinch, S., Alt, F.: Pervasive displays: Understanding the future of digital signage, Synthesis Lectures on Real-Time Systems, 9th International Workshop on Programming Models and Applications for Multicores and Manyco- Mobile and Pervasive Computing 8 (1), 2014 res (PMAM), 2018 03. Buschek, D., De Luca, A., Alt, F.: Improving Accuracy, Applicability and Usability of Keystroke Biometrics on Mobile 02. Wägemann, T., Tavakoli Kolagari, R., Schmid, K.: Optimal Product Line Architectures for the Automotive Industry, Touchscreen Devices, 33rd CHI Conference on Human Factor in Computing Systems, 2015 S. 119ff, In: Modellierung, Bonn: Gesellschaft für Informatik e.V., 2018 03. Tavakoli Kolagari, R., Chen D., Lanusse, A., Librino, R., et al.: Model-based Analysis & Engineering of Automotive Architectures #Mensch-Maschine Interaktion #Ubiquitous Computing #Benutzbare Sicherheit with EAST-ADL: Revisited, International Journal of Conceptual Structures and Smart Applications (IJCSSA), 2016

#Autonomes Fahren #modellbasierte Softwareentwicklung #Automotive Software Systems Engineering #Software Architektur #Requirements Engineering

22 | AKTEURE AKTEURE | 23 AKTEURE AKTEURE

ZD.B Nachwuchs- ZD.B Nachwuchs- forschungsgruppe forschungsgruppe

Dr. Daniel Schnurr Dr.-Ing. Janet Siegmund Data Neutrality & Open Access: Coherent PICCARD: Improving Collaboration Economic Policies for the Digital Economy and Communication of Developers Universität Passau Universität Passau www.datapolicies.net https://www.infosun.fim.uni-passau.de/se/people-jsiegmund.php

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Im Fokus unserer Forschung stehen Daten und deren zentrale es Maßnahmen zu identifizieren, die geeignet sind langfristig Software findet sich heutzutage überall: Im Auto, in der Softwareprojekts, sodass wir diese beiden Aspekte im Detail Rolle für den Wettbewerb in digitalen Märkten. Insbesondere effektiven Wettbewerb zu sichern und Innovation zu fördern. Waschmaschine, sogar im Kühlschrank. Der Faktor Mensch untersuchen. Das Ziel ist dabei, Kommunikation und Kolla- beschäftigen wir uns mit der Frage, wie der Zugang zu Daten Mit Hilfe spieltheoretischer Modelle, ökonomischer Labor- spielt eine wesentliche Rolle beim Entwickeln und Warten boration zu verbessern, damit entsprechend individueller ausgestaltet und geregelt werden sollte. Diese Frage stellt experimente und Simulationen untersuchen wir (Data) Open dieser Software, sodass er den Fokus dieses Projekts bildet. Faktoren von Entwicklern und Eigenschaften von Software- sich sowohl für Unternehmen, um im Wettbewerb bestehen Access Konzepte, Neutralitätsverpflichtungen für Unterneh- Insbesondere ist die Kommunikation und Kollaboration ein projekten eine optimale Strategie angewendet werden kann. zu können, als auch auf gesellschaftlicher Ebene mit Blick auf men sowie Maßnahmen zur Stärkung der informationellen entscheidender Faktor für den Erfolg oder Misserfolg eines den gesamtwirtschaftlichen Ordnungsrahmen. Unser Ziel ist Selbstbestimmung. KURZER LEBENSLAUF KURZER LEBENSLAUF Janet Siegmund leitet derzeit die Nachwuchsforschungsgrup- hat mehr als 30 begutachtete Artikel in Journalen, Konferen- Daniel Schnurr leitet seit April 2017 die Nachwuchsforschungs- Marketing (IISM) arbeitete. Von 2007 bis 2013 studierte Daniel pe PICCARD, gefördert durch das ZD.B, an der Universität zen und Workshops veröffentlicht. Sie ist regelmäßig Mitglied gruppe Data Policies an der Universität Passau. Er promovierte Schnurr Informationswirtschaft (B.Sc. & M.Sc.) am Karlsruher Passau. Sie promovierte 2012 an der Universität Magdeburg in Programmkommittees von Konferenzen und Workshops. 2016 im Bereich Wirtschaftschaftsinformatik am Karlsruher Institut für Technologie. Er war Gaststudent an der John Mol- und ist sowohl Diplom-Psychologin als auch Diplom-Informa- Von 2014 bis 2017 war sie im Steering-Committee der Inter- Institut für Technologie, wo er drei Jahre als wissenschaftli- son School of Business, Concordia University (Kanada) und der tikerin. In ihrer Forschung beschäftigt sie sich mit der Rolle national Conference on Program Comprehension. cher Mitarbeiter am Institut für Informationswirtschaft und Singapore Management University (Singapur). des Faktor Mensch im Software-Engineering. Janet Siegmund

AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN

01. Krämer, J., Schnurr, D., Wohlfarth, M.: Winners, Losers, and Facebook: The Role of Social Logins in the Online Advertising 01. Peitek, N., Siegmund, J., Apel, S., Kästner, et al.: A Look into Programmers' Heads, IEEE Transactions on Software Ecosystem, Management Science, 2018 Engineering (TSE), 2018 02. Krämer, J., Schnurr, D.: Is There a Need for Platform Neutrality Regulation in the EU? Telecommunications Policy, 42(7), 2018 02. Siegmund, J., Siegmund, N., Apel, S.: Views on Internal and External Validity in Empirical Software Engineering, 03. Horstmann, N., Krämer, J., Schnurr, D.: Number Effects and Tacit Collusion in Experimental Oligopolies, Journal of IEEE/ACM International Conference on Software Engineering (ICSE), 2015 Industrial Economics, 2018 03. Joblin, M., Mauerer, W., Apel, S., Siegmund, J., Riehle, D.: From Developer Networks to Verified Communities: A Fine-Grained Approach, IEEE/ACM International Conference on Software Engineering (ICSE), 2015 #Wettbewerb in digitalen Märkten #Zugang zu Daten #Informationelle Selbstbestimmung #Economics of ICT Softwareentwicklung, Entwicklerkoordination, Netzwerkanalyse

24 | AKTEURE AKTEURE | 25 AKTEURE AKTEURE

ZD.B Nachwuchs- ZD.B Nachwuchs- forschungsgruppe forschungsgruppe

Dr. Matthias Uhl Dr. Christian Wachinger Ethik der Digitalisierung Computational Population Modeling Technische Universität München from Big Medical Image Data https://www.professoren.tum.de/en/tum-junior-fellows/ u/dr-matthias-uhl/ Ludwig-Maximilians-Universität München www.ai-med.de

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Die Forschungsgruppe entwickelt konkrete Politikvorschläge gesellschaftliche Widerstände gegen die Digitalisierung, die Künstliche Intelligenz verändert das Gesundheitswesen. gelegte Forschungsstudien und jahrzehntelange Bildgebung auf organisatorischer und staatlicher Ebene, die helfen, die entweder durch reflektierte Urteile oder unbewusste Intui- Insbesondere medizinische Bilder enthalten umfangreiche in der Klinik riesige Bildarchive. Die Analyse solch riesiger vielfältigen ethischen Herausforderungen der Digitalisierung tionen bezüglich des Einsatzes von künstlicher Intelligenz in Informationen, die möglicherweise nur teilweise mit blo- Datensätze eröffnet viele spannende Forschungsfragen, so- zu bewältigen. Ein erster Forschungsstrang widmet sich der neuen Domänen verursacht werden. Die Gruppe verwendet ßem Auge erkennbar sind, aber von intelligenten Algorith- wohl aus klinischer als auch aus technischer Sicht. In enger Frage, wie Mensch-Maschine-Interaktionen in einer Welt ge- empirische Methoden, insbesondere ökonomische Experi- men identifiziert werden können. Für das Trainieren solcher Zusammenarbeit mit unseren klinischen Partnern arbeiten staltet werden sollen, in der zahlreiche Entscheidungen mit mente, um moralisches Verhalten direkt zu beobachten, Methoden sind allerdings große Datenmengen erforderlich. wir an intelligenten Algorithmen um diese Fragen zu beant- moralischen Konsequenzen im digitalen Umfeld getroffen anstatt sich ausschließlich auf hypothetische Aussagen zu Gleichzeitig entstanden in den letzten Jahren durch groß an- worten. werden. Ein zweiter Forschungsstrang konzentriert sich auf verlassen. KURZER LEBENSLAUF KURZER LEBENSLAUF Christian Wachinger, Jahrgang 1982, hat an der TU München Computer Science and Artificial Intelligence Lab, MIT und Matthias Uhl studierte zunächst Betriebswirtschaftslehre an der Jena. Von März 2012 bis Januar 2017 war er Postdoc am Pe- Informatik studiert und dort bei Prof. Navab promoviert. am Department of Neurology, Harvard Medical School. Von Fachhochschule Mainz (2000-2004) und anschließend Volks- ter-Löscher-Lehrstuhl für Wirtschaftsethik der TU München. Er hat studienbegleitend das Center for Digital Technolo- 2015 bis 2018 vertrat er die Professur für neurobiologische wirtschaftslehre an der Johannes Gutenberg-Universität Mainz Seit Februar 2017 ist er Leiter der Nachwuchsgruppe "Ethik gy and Management besucht und Auslandsstudien an der Forschung am Klinikum der Universität München. Seit 2017 (2004-2007). Anschließend forschte er am Max-Planck-Institut der Digitalisierung" an der TUM. Weitere Forschungsinteressen Ecole Nationale Superieure des Telecommunications und führt er die ZD.B Nachwuchsgruppe. für Ökonomik in Jena und promovierte 2011 mit einer verhal- liegen in den Bereichen der experimentellen Ökonomik, der der Princeton University absolviert. Als PostDoc war er am tensökonomischen Arbeit an der Friedrich-Schiller-Universität Wirtschaftsethik sowie der Wissenschaftstheorie. AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN 01. Wachinger, C., Nho, K., Saykin, A., Reuter, M., Rieckmann, A.: A Longitudinal Imaging Genetics Study of Neuroanatomical 01. Gogoll, J., Uhl, M.: Rage Against the Machine: Automation in the Moral Domain, Journal of Behavioral and Experimental Asymmetry in Alzheimer’s Disease, Biological Psychiatry, 2018 Economics, 2018 02. Wachinger, C., Salat, D., Weiner, M., Reuter, M.: Whole-Brain Analysis Reveals Increased Neuroanatomical Asymmetries 02. Ostermaier, A., Uhl, M.: Spot on for Liars! How Public Scrutiny Influences Ethical Behavior, PLOS ONE, 2017 in Dementia for Hippocampus and Amygdala, Brain, 2016 03. Jauernig, J., Uhl, M., Luetge, C.: Competition-induced Punishment of Winners and Losers: Who is the Target?, 03. Wachinger, C. Navab, N.: Simultaneous Registration of Multiple Images: Similarity Metrics and Efficient Optimization, Journal of Economic Psychology, 2016 IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 2013

#Moralisches Handeln #Mensch-Maschine-Interaktionen #Algorithmenethik #Akzeptanzforschung #Medical Imaging #AI #Neuro #Big Data #Experimentelle Ethik

26 | AKTEURE AKTEURE | 27 AKTEURE AKTEURE

ZD.B Nachwuchs- forschungsgruppe ZD.B Fellow

Dr. Raphael Wimmer Alice Baird Physical Affordances for Digital Media Ein Embedded-Soundscape-System für and Workflows personalisiertes Wohlbefinden durch Universität Regensburg multimodale Feedbackanalyse https://hci.ur.de Universität Augsburg https://www.informatik.uni-augsburg.de/lehrstuehle/eihw/staff/baird/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Die Digitalisierung verändert die Arbeit mit Information. besitzen, und wie ihre jeweiligen Stärken in Arbeitsprozessen Die „Soundscape“ (die hörbaren Komponenten einer be- werden die Vorteile synthetischen Soundscape Designs er- Digitale Systeme und Werkzeuge ersetzen Papier, Stift und sinnvoll kombiniert werden können. Neben der Forschung stimmten Umgebung) ist eine Allgegenwart im täglichen Le- forscht. Personalisierung der Soundscape erfolgt im Rahmen Stempel. Hier müssen Entwickler, Designer und Entscheider an geeigneten Hardware- und Software-Architekturen unter- ben. Die Forschung hat jedoch gezeigt, dass Elemente unserer des Einsatzes von Verfahren tiefen Lernens in eingebetteten darauf achten, dass die ganz eigenen Vorteile papierbasierter suchen wir auch in Fallstudien, wie traditionell papierbasierte akustischen Geräuschkulissen psychologisches Wohlbefinden Systemen zur Verbesserung des menschlichen Wohlbefindens kollaborativer Prozesse auch in digitalen Lösungen erhalten Arbeitsprozesse möglichst benutzerfreundlich digital umge- negativ beeinflussen können. Im Rahmen eines dualen Ana- im Alltag. Das Projekt wird Fragestellungen der akustischen bleiben. In der ZD.B-Nachwuchsgrupppe „Physical-Digital setzt werden können. Ein besonderer Fokus liegt hierbei auf lyse-Synthese- Ansatzes wird eine technische Anpassung Wahrnehmung und zur Untersuchung entwickelter Methoden Affordances“ untersuchen wir, welche inhärenten Nutzung- der Arbeit in der öffentlichen Verwaltung – einem Sektor, der dieser angestrebt. Mittels multimodaler Feedback-Analyse zur intelligenten Erstellung von Soundscapes erforschen. möglichkeiten („Affordances“) physische und digitale Medien stark von papierbasierten Prozessen geprägt ist. KURZER LEBENSLAUF KURZER LEBENSLAUF Alice Baird (geb. 1992 im Vereinigten Königreich), ist ein Fel- terin an der Universität Passau tätig. Ihr interdisziplinärer Raphael Wimmer, geboren 1980, hat an der Ludwig-Maximili- Mensch-Computer-Interaktion an der Schnittstelle zwischen low des ZD.B, betreut von Professor Dr. habil. Björn Schuller Forschungsschwerpunkt liegt in der intelligenten allgemei- ans-Universität München Medieninformatik studiert und dort Computer-Hardware und menschlichen Sinnen. Seit April am ZD.B Lehrstuhl für Embedded Intelligence for Health nen Audioanalyse. 2013 erhielt Alice einen BA der London 2015 bei Prof. Dr. Heinrich Hußmann über Griffmustererken- 2017 leitet er die ZD.B-Nachwuchsgruppe „Physical-Digital Care and Wellbeing der Universität Augsburg. Zuvor war Metropolitan Universität und 2016 einen MFA von der Co- nung für die Mensch-Computer-Interaktion promoviert. Seit Affordances“ am Institut für Information und Medien, Spra- sie als wissenschaftliche Mitarbeiterin für das EU-Horizon lumbia University, New York City. Alice hat seit 2016 10+ 2011 lehrt und forscht er an der Universität Regensburg zur che und Kultur (I:IMSK) der Universität Regensburg. 2020-Projekt DE-ENIGMA als wissenschaftliche Mitarbei- Co-Autorenschaften zu diesem Thema.

AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN

01. Wimmer, R., Hennecke, F., Schulz, F., Boring, S., Butz, A., Hußmann, H.: Curve: revisiting the digital desk, 01. Cummins, N., Baird, A., Schuller B.: The increasing impact of deep learning on speech analysis for health: 6th Nordic Conference on Human-Computer Interaction: Extending Boundaries (NordiCHI), 2010 Challenges and Opportunities, Methods – Special Issue on Translational Data Analytics and Health Informatics, 2018 02. Grosse-Puppendahl, T., Holz, C., Cohn, G., Wimmer, R., Bechtold, O., Hodges, S., Reynolds, M.S., Smith, J.R.: Finding 02. Baird, A., Hantke, S., Schuller, B.: Responsible and Representative Multimodal Data Aquisistion: On Auditability, Common Ground: A Survey of Capacitive Sensing in Human-Computer Interaction, 2017 CHI Conference on Human Benchmarking, Confidence, Data-Reliance & Explainability, Language Resource and Evaluation Conference (LREC), 2018 Factors in Computing Systems (CHI), 2017 03. Baird, A., Parada-Cabaleiro, E., Fraser, C., Hantke, S., Schuller, B.: The Perceieved Emotion of Isolated Synthetic Audio – 03. Wimmer, R., Hahn, J.: A Concept for Sketchable Workspaces and Workflows, Workshop "Rethinking Interaction" in A Dataset and Results, Audio Mostly, 2018 Conjunction with CHI, 2018 #Computer Audition #Wellbeing #Machine Learning #Mensch-Computer-Interaktion #Digitale Verwaltung #Tangible Interaction #Affordances

28 | AKTEURE AKTEURE | 29 AKTEURE AKTEURE

ZD.B Fellow ZD.B Fellow

Clemens Birkenmaier Philipp Dufter Realistische Modellierung von Blut in Effektive Nutzung mehrsprachiger Daten patientenspezifischen digitalen Modellen in der automatisierten Sprachverarbeitung Universität Regensburg, OTH Regensburg Ludwig-Maximilians-Universität München https://bfm.rcbe.de/ http://www.cis.lmu.de

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Warum gerinnt Blut in medizintechnischen Geräten wie Strömungs- und Gerinnungsverhalten von Blut in künstlichen Zahlreiche Anwendungen in der Sprachverarbeitung benöti- in einer Sprache zu reduzieren indem Daten über mehrere Dialysatoren oder künstlichen Lungen, obwohl es nach bis- Lungen zu entwickeln. Das Projekt wird die Entwicklung besse- gen große Mengen an – meist manuell – annotierten Daten. Sprachen hinweg effizient genutzt werden. In der langen Frist herigem Verständnis nicht sollte? Ziel ist es, die komplexen rer Medizinprodukte mittels virtueller Prototypen ermöglichen Derzeit betrachtet die Vielzahl der verwendeten Methoden können so Methoden nicht nur auf eine größere Anzahl von Zusammenhänge hinter dieser Frage zu verstehen und ein und neue Wege bei Diagnose und Therapieplanung für die lediglich eine oder zwei Sprachen gleichzeitig, obwohl die Sprachen (in denen wenig Daten vorhanden sind), sondern patientenindividuelles Simulationsmodell für das kombinierte digitale Präzisionsmedizin der Zukunft eröffnen. Daten oft in einer Vielzahl an Sprachen vorliegen. Dieses For- auch auf spezialisierte Domänen, die ebenfalls oft einen Man- schungsprojekt hat zum Ziel, die erforderliche Datenmenge gel an brauchbaren Daten aufweisen, angewendet werden. KURZER LEBENSLAUF KURZER LEBENSLAUF Clemens Birkenmaier, geboren 1989, studierte Maschinen- terarbeit im Bereich der numerischen Strömungsrechnung bau mit Schwerpunkt numerische Methoden in Konstanz und (ausgezeichnet mit dem Preis der Freunde der OTH Regens- Philipp studierte Mathematik an der Technischen Universität BMW zahlreiche Algorithmen implementieren und so Erfahrun- Regensburg. Stipendien der Studienstiftung des deutschen burg) begann er 2016 als wissenschaftlicher Mitarbeiter im München mit einem Fokus auf Wahrscheinlichkeitstheorie. Bei gen in Computer Science sammeln. Seit einem Jahr arbeitet er Volkes und der Dr. Jürgen und Irmgard Ulderup-Stiftung Forschungsgebiet Biofluidmechanik bei Prof. Dr. Krenkel an einem anschließenden Master am Imperial College London be- mit Methoden des Maschinellen Lernens in der Sprachverarbei- ermöglichten unter anderem Praktika in der industriellen der OTH Regensburg mit Fokus auf digitalen Blut- und At- schäftigte er sich vorrangig mit stochastischer und robuster tung, ein Forschungsgebiet in dem die Themen Wahrschein- Forschung sowie einen Studien- und Forschungsaufenthalt mungsmodellen. Seit Januar 2018 ist er ZD.B-Fellow an der Optimierung. Neben dem Studium konnte Philipp als Arbeit- lichkeitstheorie, Optimierung und Computer Science mit einer an der Linköpings Universitet (Schweden). Nach der Mas- Universität und OTH Regensburg. nehmer in der Industrie unter anderem bei Dunnhumby und hochaktuellen Anwendung kombiniert werden.

AUSGEWÄHLTE PUBLIKATIONEN AUSGEWÄHLTE PUBLIKATIONEN

01. Birkenmaier, C., Lehle, K., Krenkel, L.: Towards a realistic model of blood viscosity and coagulation in membrane 01. Dufter, P., et al.: Embedding Learning Through Multilingual Concept Induction, 56th Annual Meeting of the Association oxygenators, 7th European Conference on Computational Fluid Dynamics (ECCOMAS ECFD7), 2018 for Computational Linguistics, 2018 02. Birkenmaier, C., Philipp, A., Müller, T., Gruber, M., et al.: Detection and analysis of clot deposits in extracorporeal membrane oxygenators using high-resolution computed tomography. Perfusion, 33(1), 2018 #Natural Language Processing #Machine Learning #Deep Learning #Multilinguality 03. Birkenmaier, C., Krenkel, L.; Lehle, K.: Linking flow conditions in membrane oxygenators to arrangements of multimeric von-Willebrand-factor as indication for coagulation, 8th World Congress of Biomechanics (WCB2018), 2018

#Präzisionsmedizin #simulation-based decision support #inter-scale modelling

30 | AKTEURE AKTEURE | 31 AKTEURE AKTEURE

ZD.B Fellow ZD.B Fellow

Oliver Haas Christoph Molnar Digitalisierung im klinischen Umfeld Interpretierbares maschinelles Lernen mittels Graphdatenbanken Ludwig-Maximilians-Universität München http://www.compstat.statistik.uni-muenchen.de/people/molnar/ FAU Erlangen-Nürnberg, OTH Amberg-Weiden https://www.oth-aw.de/informieren-und-entdecken/fakultaeten/ wirtschaftsingenieurwesen/ueber-die-fakultaet-wirtschaftsingenieurwesen/

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Im klinischen Umfeld werden an verschiedenen Stellen Da- werden können, vor allem wie die Mitarbeiter der Klinik Maschinelles Lernen ermöglicht es Computern, selbstständig gen als auch das Gesamtverhalten eines Vorhersagemodells ten erhoben und gesammelt. Diese sind stark heterogen, davon profitieren können. Denkbare Szenarien sind eine aus Daten zu lernen, um Vorhersagen zu treffen, zum Bei- zu interpretieren. Wir entwickeln die Methoden mathema- aber dennoch hochgradig verknüpft. Graphdatenbanken graphische Oberfläche, damit Mitarbeiter einfacher an be- spiel automatische Diagnose basierend auf Patientendaten tisch und setzen sie in Software um. Die Herausforderung eignen sich gut für solche Daten: sie beinhalten heterogene nötigte Informationen kommen oder eine Schnittstelle für oder Bestimmung der Kreditwürdigkeit basierend auf Kredit- besteht darin, komplexe Vorhersagemodelle auf Kennzahlen „Datenpunkte“ die miteinander verknüpft werden können. Auswertungen, um komplexere Fragen beantworten. (z. B. ausfalldaten. Die erlernten Entscheidungsmechanismen sind (z.B. die Wichtigkeit einzelner Eingaben) zu reduzieren und Die grundlegende Fragestellung ist nun, ob und wie diese „Gab es zu einem Patienten bereits ähnliche Patienten? Wie meist nicht nachvollziehbar. Im Forschungsgebiet für inter- für den Menschen verständliche Erklärungen zu liefern, ohne klinischen Daten in einer Graphdatenbank sinnvoll genutzt wurden sie behandelt?“) pretierbares maschinelles Lernen entwickeln wir Methoden, Vorhersagemodelle zu stark zu vereinfachen. die es Menschen ermöglichen, sowohl individuelle Vorhersa- KURZER LEBENSLAUF KURZER LEBENSLAUF Geboren in Nürnberg und aufgewachsen in Fürth, begann er für drei Jahre in der Softwarebranche tätig, wo er mit der die akademische Karriere von Oliver Haas 2010 an der FAU Medizintechnik in Berührung kam und viel über Software Christoph Molnar hat an der LMU München Statistik studiert. zu machen, führte ihn dazu ein Buch darüber zu beginnen Erlangen-Nürnberg. Sein Bachelor- und Masterstudium in Engineering und digitale Prozessketten im klinischen Umfeld Nach dem Studium ist er in die Schweiz ausgewandert, wo er und nach München zurückzukehren, um tiefer in das Thema Mathematik (mit den Nebenfächern Theoretische Physik und lernte, was schließlich zum Thema seiner Forschung geführt drei Jahre lang in der Wirtschaft tätig war, zunächst bei einem einzutauchen. Seit Oktober 2017 promoviert Christoph bei Informatik) schloss er 2015 ab, wobei der Schwerpunkt in hat. Seit Oktober 2018 wird die Promotion durch ein ZD.B Start-up im Bankwesen und später in der medizinischen For- Prof. Bischl am Institut für Statistik der LMU München auf der Algebraischen Topologie lag. Nach seinem Studium war Fellowship gefördert. schung. Sein Interesse, maschinelles Lernen interpretierbar dem Gebiet des interpretierbaren maschinellen Lernens.

#Digitalisierung #Graphdatenbanken #klinischer Workflow #digitale Medizin AUSGEWÄHLTE PUBLIKATIONEN

01. Molnar, C.: Interpretable Machine Learning, 2018 (Retrieved from https://christophm.github.io/interpretable-ml-book/) 02. Molnar, C., Casalicchio, G., Bischl, B.: iml: An R package for Interpretable Machine Learning . Journal of Open Source Software, 2018 03. Casalicchio, G., Molnar, C., Bischl, B.: Visualizing the Feature Importance for Black Box Models, 2018

#Maschinelles Lernen #Künstliche Intelligenz #Algorithmus

32 | AKTEURE AKTEURE | 33 AKTEURE AKTEURE

ZD.B Fellow ZD.B Fellow

Nina Rohrbach Jonas Schlund „Mixed Reality” als ein neuer rehabilitativer Blockchain-basierte Orchestrierung von Ansatz bei Störungen von Alltagshandlungen verteilten Assets im zukünftigen Stromsystem nach chronisch neurologischer Erkrankung Friedrich-Alexander-Universität Erlangen-Nürnberg http://www7.cs.fau.de/de/ Technische Universität München https://www.bewegungswissenschaft.sg.tum.de

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Die therapeutischen Möglichkeiten im Bereich der Neuro- zur selbstbestimmten Teilhabe und somit zu höherer Le- Ziel des Promotionsvorhabens ist es zu erforschen, wie ver- Ansätze zur Orchestrierung, die auf der Kombination von rehabilitation sind vielseitig. Hierbei begeistert mich insbe- bensqualität auch nach chronischer Erkrankung beizutragen. teilte Assets in einem zukünftigen Stromnetz, welches auf Blockchain-Technologie und dem zellularen Ansatz basieren, sondere die Erforschung innovativer Rehabilitationsansätze Ich bin davon überzeugt, dass die gewinnbringende Entwick- erneuerbaren Energien und Speichern basiert, orchestriert werden herkömmlichen zentralen Lösungen gegenüberge- zur Optimierung der motorischen Kontrolle nach Schädigung lung digitaler Lösungen und v.a. die Integration dieser in die werden können. Insbesondere auf Verteilnetzebene werden stellt. Zudem wird ein quantifizierbares Bewertungsraster zur des Nervensystems, z.B. mittels Robotik oder Mixed-Reali- klinische Routine nur im transdisziplinären Team und in Zu- immer mehr verteilte und heterogene Erzeuger Erneuerba- Evaluation des ganzheitlichen Mehraufwandes der verteilten ty-Technologie. Die Entwicklung technischer Assistenzsys- sammenarbeit aller Stakeholder gelingen kann, weshalb ich rer Energien sowie Speicher von Privatpersonen oder Fir- Ansätze entwickelt und dieses für verschiedene Blockchains teme zur Unterstützung von Alltagsaktivitäten ist komplex, den partizipativen Forschungsdesigns einen großen Stellen- men installiert, die nicht zentral gesteuert werden. Verteilte und die vielversprechendsten Anwendungsfälle angewendet. doch haben solche Systeme das enorme Potential nachhaltig wert zuweise. KURZER LEBENSLAUF KURZER LEBENSLAUF Nach seinem Abitur 2010 studierte Jonas Schlund Energie- für Angewandte Mathematik 3 in Erlangen, im Centro Técnico Nina Rohrbach hat aufbauend auf ihr Staatsexamen an der an der FAU Erlangen zum Thema der Verhaltensänderung technik in Erlangen. Für seine Bachelorarbeit 2014 erhielt de Seat in Martorell sowie bei Siemens in Erlangen und Nürn- HS Osnabrück und der Curtin University Perth, Australien, zur Steigerung körperlicher Aktivität. Seit 2017 promoviert er den Brose-Bachelorpreis. Sein Masterabschluss mit Aus- berg tätig. Nach dem Studium entschied er sich als Doktorand Physiotherapie studiert (B.Sc.). Nach klinischer Praxiserfah- Nina am Lehrstuhl für Bewegungswissenschaft der TUM bei zeichnung erfolgte 2016 im Studienschwerpunkt "elektrische an der Universität eine Promotion im Bereich der vernetzten rung in verschiedenen Einrichtungen (Schwerpunkt Neuro- Prof. Dr. Joachim Hermsdörfer zum Einsatz von Mixed Reality Energietechnik". Während des Studiums war er am Lehrstuhl Energiesysteme anzustreben. rehabilitation) hat sie ihren M.Sc. (mit Auszeichnung) in Be- Technologie in der Neurorehabilitation. Als Visiting Scholar wegung und Gesundheit an der TUM abgeschlossen. Für ihre forschte sie kürzlich im Rehabilitation Games & Virtual Reality Masterarbeit in Kooperation mit der ETH Zürich erhielt sie Lab der Northeastern University in Boston, USA. AUSGEWÄHLTE PUBLIKATIONEN den Dr. Gertrude Krombholz Preis. Anschließend forschte sie 01. Schlund, J., Ammon, L., German, R.: ETHome: Open-source Blockchain Based Energy Community Controller, 9th Interna- tional Conference on Future Energy Systems, 2018 AUSGEWÄHLTE PUBLIKATIONEN 02. Schlund, J., Steinert, R., Pruckner, M.: Coordinating E-Mobility Charging for Frequency Containment Reserve Power Provi- sion, 9th International Conference on Future Energy Systems, 2018 01. Rohrbach, N., Gulde, P., Armstrong, A., et al.: An augmented reality approach for ADL support in Alzheimer's disease: 03. Schlund, J., Betzin, C., Wolfschmidt, H., et al.: Investigation, modeling and simulation of redox-flow, lithium-ion and lead-acid a crossover trial, 2018 battery systems for home storage applications, 11th International Renewable Energy Storage Conference (IRES), 2017 02. Baur K., Rohrbach N., Hermsdörfer J., et al.: "Beam-Me-In" Strategy – Remote Haptic Therapist-Patient Interaction with Two Exoskeletons for Stroke Therapy, 2018 #Dezentrale Energieversorgung #Blockchain #Zellularer Ansatz #Erneuerbare Energien 03. Buckingham G., Bieńkiewicz M., Rohrbach N., Hermsdörfer J.: The impact of unilateral brain damage on weight perception, sensorimotor anticipation, and fingertip force adaptation, Vision research, 2015

#Neurorehabilitation #Mixed Reality Technologie #Alltagshandlung #Motorisches Lernen #Assistenzsysteme

34 | AKTEURE AKTEURE | 35 AKTEURE AKTEURE

ZD.B Fellow ZD.B Fellow

Verena Süß Dipl.-Ing. Johanna Wald Selbstregulierung im 3D Szenenverständnis von europäischen Datenschutzrecht statischen und dynamischen Umgebungen Universität Bayreuth Technische Hochschule München campar.in.tum.de/Main/JohannaWald

FORSCHUNGSSCHWERPUNKTE FORSCHUNGSSCHWERPUNKTE

Die Digitalisierung stellt den Datenschutz vor neue Heraus- zusetzen. Es verspricht unter anderem, die Akzeptanz bei Um effektive und intuitive Interaktion mit Smart Devices zu nen z. B. der Pose, zugehöriger Objektkategorien sowie der forderungen. Wie kann angemessen auf die veränderte Ge- den Datenschutzverpflichteten zu erhöhen. Die Datenschutz- ermöglichen sind Algorithmen zur Verarbeitung visueller Bestimmung der räumlichen, semantischen und temporalen fährdungslage reagiert werden? Als Antwort auf die beobach- grundverordnung sieht mit Verhaltensregeln, Zertifizierungen Umgebungsinformationen notwendig. Der Fokus unserer Beziehungen der Objekte. Auch die Repräsentation der 3D teten Defizite des herkömmlichen Datenschutzrechts wird das und verbindlichen internen Datenschutzvorschriften (Art. 40f., Forschung liegt deshalb auf der Entwicklung von Methoden Daten spielt dabei in der Forschungsfrage eine wichtige Rolle Steuerungskonzept der „Regulierten Selbstregulierung“ disku- 42f., 47 DS-GVO) offenbar entsprechende Instrumente vor. In- zum visuellen Verständnis von statischen und dynamischen um Effizienz und gleichzeitig gute Ergebnisse zu erhalten. tiert. Nach diesem Konzept soll die Wirtschaft dazu veranlasst wieweit der europäische Datenschutz durch diese Instrumen- 3D Szenen. Dazu zählt die Extraktion von Objektinformatio- werden, branchen- oder unternehmensbezogene Regelungen te tatsächlich weiterentwickelt werden kann, wird im Rahmen zu entwerfen und mithilfe privater Kontrollinstanzen durch- des Projektes untersucht. KURZER LEBENSLAUF

KURZER LEBENSLAUF Johanna ist eine Doktorandin der TUM, die an 3D Szenen- ihren Master, ihre Masterarbeit Real Time Incremental Scene verständnis arbeitet. Ihre Forschungsinteressen liegen im Understanding on Google Tango wurde in Kooperation mit Verena Süß, geboren 1993, hat an der Universität Bayreuth abteilung einer internationalen Großkanzlei promoviert sie Bereich (Echtzeit) Computer Vision Algorithmen. Durch die Google München und dem CV Team am CAMP Lehrstuhl der von 2012–2017 Rechtswissenschaft mit wirtschaftswissen- seit Dezember 2017 zu dem Arbeitstitel „Selbstregulierung Wahl ihres Bachelors, mit Fokus auf Computergrafik und AR, TUM erstellt. Ihren PhD begann sie im September 2017 und schaftlicher Zusatzausbildung studiert. Im Termin 2017/I hat im europäischen Datenschutzrecht“. Seit Januar 2018 wird konnte sie sich bereits früh spezialisieren. 2017 beendete sie ist seit 2018 ein ZD.B Fellow. sie die Erste Juristische Prüfung abgelegt. Nach einer Tätigkeit die Promotion durch ein ZD.B Fellowship gefördert. als wissenschaftliche Mitarbeiterin in der Datenschutzrecht- AUSGEWÄHLTE PUBLIKATIONEN

#Datenschutzzertifizierung #Codes of Conduct #Binding Corporate Rules #Selbstregulierung #Selbstkontrolle 01. Rethage, D., Wald, J., Sturm, J., et al.: Fully-Convolutional Point Networks for Large-Scale Point Clouds, 15th European Conference on Computer Vision (ECCV), 2018 02. Wald, J., Tateno, K., Sturm, J., et al.: Real-Time Fully Incremental Scene Understanding on Mobile Platforms, IEEE Robotics and Automation Letters (RA-L), 2018

#3D Szenenverständnis #Semantische Segmentierung #Maschinelles Sehen #Robotik #Augmented Reality

36 | AKTEURE AKTEURE | 37 AKTEURE

ZD.B Fellow

Sarah Wunderlich Selbstadaptierende Online-Analyse hostbasierter sicherheitskritischer Ereignisse in Unternehmens- netzwerken kleiner und mittlerer Unternehmen Hochschule Coburg, Universität Würzburg https://www.hs-coburg.de/CIDDS

FORSCHUNGSSCHWERPUNKTE

Mit der stärkeren Digitalisierung von Unternehmensprozes- Projekts ist es ein selbstlernendes und benutzerfreundliches sen steigen auch Umfang und Qualität der zu schützenden System zu entwerfen, welches relevante host-basierte Daten Daten, während zugleich eine deutlich größere Angriffsfläche sammelt und analysiert, um Angriffe effektiv entdecken und für Angreifer geboten wird. Komplexe Sicherheitslösungen eindämmen zu können. Zudem sollen die Ergebnisse der wie Intrusion Detection Systeme (IDS) finden jedoch in klei- Analyse so aufbereitet werden, dass sich das System für nen und mittelständischen Unternehmen (KMUs) aufgrund KMUs mit geringen Personalressourcen und Spezialwissen von fehlendem Know-how häufig keine Anwendung. Ziel des eignet.

KURZER LEBENSLAUF

Sarah Wunderlich hat an der Hochschule Coburg Informatik gen, Technikbegeisterung in Kindern zu erwecken, wurden studiert und arbeitet dort nun als wissenschaftliche Mit- 2015 mit einem Förderpreis des Stifterverbandes für die arbeiterin in der Arbeitsgruppe von Professor Dr. Dieter Deutsche Wissenschaft ausgezeichnet. Ebenfalls war sie Landes im Bereich IT Sicherheit durch Maschinelles Lernen. bereits als Lehrbeauftragte für das Modul "Data Mining" an Während des Studiums war sie nebenbei am Fraunhofer der Hochschule tätig. Seit Januar 2018 ist sie Empfängerin IIS tätig und erhielt mehrere Stipendien. Ihre Bemühun- eines ZD.B Fellowships.

AUSGEWÄHLTE PUBLIKATIONEN

01. Wunderlich, S., Schmölz, J., Kühnlenz, K.: Follow me: A simple approach for person identification and tracking, 26th IEEE Symposium on Industrial Electronics (ISIE), 2017 02. Ring, M., Wunderlich, S., Grüdl, D., et al.: Flow-based benchmark data sets for intrusion detection, 16th European Conference on Cyber Warfare and Security (ECCWS), 2017 03. Ring, M., Wunderlich, S., Grüdl, D., et al.: A toolset for intrusion and insider threat detection, In: Data Analytics and Decision Support for Cybersecurity, Springer, 2017

#Datenschutzzertifizierung #Codes of Conduct #Binding Corporate Rules #Selbstregulierung #Selbstkontrolle

38 | AKTEURE 39 ARBEITSKREISE

AKTEURE SEITE 7 ARBEITSKREISE DER INITIATIVEN FÜR DIE WISSENSCHAFT DES ZD.B

„Digitalisierung“ ist ein vielschichtiger Begriff, der Klärung und Kontext bedarf. Um Digitalisierung in ihrer Komplexität für verschiedene Interessensgruppen aufbereiten und gleichzeitig ihren Nutzen für Wissenschaft, Gesellschaft, Politik und Unternehmen deutlich machen zu können, ist eine tiefe inhaltliche Aufbereitung der vielseitigen Themenfelder notwendig. Die Befähigung unterschiedlichste interdisziplinäre Facetten zu durchdringen stellt gleichzeitig einen entscheidenden Er- folgsfaktor und starken Erkenntnisgewinn dar. Damit werden Themen der Digitalisierung letztendlich nicht mehr nur auf Basis punktueller und eindimensionaler Perspektiven betrachten, sondern können in Wechselwirkung mit allen Ebenen (Wirtschaft, Politik, Wissenschaft und Gesellschaft) reflektiert und aktiv gestaltet werden.

Vor diesem Hintergrund wurden Anfang 2018 die ZD.B Arbeitskreise ins Leben gerufen, zunächst zu den Themen: ARBEITSKREISE Digital Society: Culture and Policies SEITE 40 Human-Centred Engineering Software Engineering Artificial Intelligence Autonomous Systems Digital Health

Ziel der selbstorganisierten ZD.B Arbeitskreise ist es, die interdisziplinären Forschungskompetenzen der Akteure der Initia- tiven für die Wissenschaft im Rahmen eines regelmäßigen Austauschs zu bündeln und durch gemeinsame Veranstaltungen und Projekte die Dissemination der (angewandten) Forschungsergebnisse – zum Beispiel über die ZD.B Themenplattformen SYMPOSIUM – effizient zu gestalten. In den ZD.B Arbeitskreisen engagieren sich sowohl Wissenschaftlerinnen und Wissenschaftler der ZD.B Initiativen (Profes- SEITE 58 suren, Nachwuchsforschungsgruppen und Promovierende des Graduate Program) als auch Expertinnen und Experten der jeweiligen Fachgebiete an den bayerischen Hochschulen und außeruniversitären Forschungseinrichtungen.

Alle o.g. Arbeitskreise haben zu ihren jeweiligen Themenfeldern Positionspapiere erstellt, die auf den folgenden Seiten des Jahresberichts dargestellt sind. In den Positionspapieren werden die konkreten Zielsetzungen der Arbeitskreise aufgezeigt und Standpunkte zu ausgewählten Themen der Digitalisierung in Form von Handlungsempfehlungen beschrieben.

40 | ARBEITSKREISE ARBEITSKREISE | 41 die genauen Auswahlkriterien vor Einsicht geschützt werden müs- Dimension und dem „Ethically Aligned Designed“ kommt eine zu- ARBEITSKREISE sen. Im Gegensatz dazu wird insbesondere im Datenschutzrecht nehmende Bedeutung zu. So wurde gezeigt, dass bereits die Bild- darauf verwiesen, dass Transparenz notwendige Voraussetzung schirmgröße die psychologische Distanz zu den Konsequenzen des für selbstbestimmte und informierte Entscheidungen in digitalen eigenen ethischen Tuns beeinflusst. Ferner stellt sich die Frage, ob Märkten sei. Mit zunehmender Automatisierung von Entscheidun- sich durch die dauerhafte Nutzung von datengetriebenen Prozes- gen gewinnt darüber hinaus nicht nur die Forderung nach reiner sen, beispielsweise in der medizinischen Diagnostik, die persönli- Transparenz, sondern das Kriterium der Erklärbarkeit oder Nach- che Gewichtung und Akzeptanz von Fehlerarten (False Positives vs. vollziehbarkeit algorithmischer Entscheidungen an gesellschaftli- False Negatives) systematisch verändert. DIGITAL SOCIETY: CULTURE AND POLICIES cher Bedeutung. Hier gilt es zu klären, welche zusätzlichen Anfor- derungen damit einhergehen. Digitalisierung und Demokratie: Die Digitalisierung stellt zugleich POSITIONSPAPIER Gefährdung und Chance für die parlamentarische, rechtsstaatlich Zurechenbarkeit und Verantwortung: Im Allgemeinen erleichtert der verfasste Demokratie dar. Eine wesentliche (kulturelle) Bedingung Einsatz von software-gesteuerten Expertensystemen es, in kom- einer vitalen Demokratie, die in der Lage ist wohlbegründete und Stetiger technischer Fortschritt mit immer kürzeren Innovations- Vordergrund, sondern insbesondere wird thematisiert, welche An- plexen Entscheidungssituationen richtig zu handeln. Das gilt für kohärente Entscheidungen zum Wohle der Bürgerschaft zu treffen, zyklen und rapide wachsende Datenmengen prägen die digitale reize und Hindernisse für die Nutzung dieser Techniken existieren Expertensysteme in der Medizin ebenso wie in den avancierten ist, dass der öffentliche Raum der Gründe (public space of reason, Gesellschaft. Um Digitalisierung für die Gesellschaft erfolgreich zu und wie das Bewusstsein der Nutzer für die Abgabe und den Wert Techniken, in den Unternehmen und Verwaltungen. Zugleich stellt public reasoning) eine permanente Begleitung und Kontrolle der gestalten, müssen geltende Regelwerke an diese Entwicklungen an- der eigenen Daten gesteigert werden kann. So stellt sich allgemei- sich die Frage nach der Verantwortung des menschlichen Akteurs. Gesetzgebung und des Regierungshandelns sicherstellt. Durch gepasst werden. Gleichzeitig müssen Politik und Gesellschaft aber ner die Frage wie die Erfahrung eines permanenten Online-Seins So ist in konkreten Anwendungsfällen zu beantworten wer für was neue digitale Kommunikationsmedien geschaffene Filterblasen auch gezielt Einfluss auf diesen Wandel nehmen. Die erhebliche die persönliche Wahrnehmung von Freiheit verändert und in wel- und in welchem Umfang eine Entscheidungsverantwortung trägt, und fragmentierte Communities dünnen den gemeinsamen Hori- Bedeutung sowie die mit der Neuheit verbundene Unsicherheit be- chen Kontexten persönliche Daten als sensibel oder weniger sensi- wenn die konkreten Einzelhandlungen zunehmend von Software- zont der Beurteilung politischer Vorhaben jedoch aus. Verstärkeref- züglich der zu regelnden Themenfelder erfordern Entscheidungen bel empfunden werden. Darüber hinaus werden die Implikationen systemen gesteuert sind. fekte (echo chambers) radikalisieren darüber hinaus die Meinungs- auf Basis wissenschaftlich fundierter Erkenntnisse. technischer Maßnahmen und rechtlicher Regelungen für die Inno- bildung. Gleichzeitig tragen digitale Foren in einem hohen Maße vationsanreize und -möglichkeiten sowie die Wettbewerbsfähigkeit Der Arbeitskreis wird sich dafür einsetzen, dass in der fachlichen, zur öffentlichen Meinungsbildung bei und können demokratischen Der ZD.B-Arbeitskreis Digital Society: Culture and Policies (kurz: DS:CP) der betroffenen Unternehmen berücksichtigt. Handlungsempfeh- öffentlichen und politischen Diskussion dieser Fragen ethische, Diskurs und Entscheidungsfindung vitalisieren. Der Arbeitskreis ersetzt sich zum Ziel, in besonders relevanten Themenfeldern Hand- lungen sollen dabei mit Blick auf die soziotechnische Praxis erarbei- technische, regulative und Nutzungsaspekte gleichermaßen be- arbeitet im Kontext dieser Phänomene Handlungsempfehlungen lungsempfehlungen für die Gestaltung der digitalen Gesellschaft auf tet werden, d.h., das vorhandene Wissen über bestehende Techno- handelt werden. In Abstimmung mit dem Arbeitskreis Artificial In- auf Grundlage wissenschaftlicher Erkenntnisse, die darauf abzielen Basis aktueller Forschung und wissenschaftlichen Diskurses zu for- logien und Institutionen sowie handelnde Akteure ist notwendiger telligence werden Überschneidungen und Unterschiede zwischen neue Formen der demokratischen Meinungsbildung erfolgreich zu mulieren. Dabei betrachtet der Arbeitskreis Digitalisierung nicht nur Ausgangspunkt für praktikable Lösungsansätze. der Frage nach der Transparenz und Erklärbarkeit von Algorithmen gestalten. als technische sondern als ethische, soziale, kulturelle und politische und der Frage nach der Nachvollziehbarkeit und Regulierung von Transformation. Individuelle Freiheit bei automatisierter Entscheidungsunterstützung: Verfahren maschinellen Lernens ausgeleuchtet. Zudem werden III.) Soziale Auswirkungen: Arbeit und Institutionen im Wandel Digitale Technologien schaffen neue Möglichkeiten individuelles mögliche Regelungen für die Zuweisung von Verantwortung und Insbesondere beschäftigt sich der Arbeitskreis mit den Auswirkun- Entscheidungsverhalten zu beeinflussen. So erlauben beispielswei- Rechenschaft erörtert und bewertet. Auswirkungen auf die Arbeitswelt und kollaboratives Arbeiten gen der Digitalisierung auf i) die individuelle Freiheit und Selbst- se Nudges eine auf die einzelne Person zugeschnittene Verhaltens- Vormals analoge Interaktionen zwischen Menschen werden heute bestimmung, ii) ethische Grundsatzfragen, iii) soziale Interaktionen beeinflussung. Diese Nudges können sozialen Zielen dienen, wenn II.) Ethische Grundsatzfragen der Digitalisierung vermehrt durch digitale Systeme ersetzt oder unterstützt. Dazu ge- und Institutionen sowie iv) den kulturtechnischen Wandel. Im Fokus sie beispielsweise umweltschädigendes Konsumverhalten reduzie- hören unter anderem die Kommunikation mit anderen Menschen, steht die Frage nach den sich ergebenden politischen Herausfor- ren. Paternalistische Ziele möchten die Person zu ihrem eigenen Die anthropologische Dimension der Digitalisierung: Ähnlich wie die die Zusammenarbeit in Teams, oder der Kontakt zu Behörden. derungen und die Erarbeitung von Handlungsoptionen, die diese Vorteil steuern. Hier könnte die Selbstbindung durch „Smart Con- Erfolge der Neurowissenschaften beeinflusst die Digitalisierung Selbstverständlich ist, dass ein verantwortungsbewusster Umgang Herausforderungen adressieren. Die interdisziplinäre Zusammen- tracts” an Bedeutung gewinnen, die der Person zu einem früheren unser Menschenbild. So besteht beispielsweise die Silicon-Val- mit digitalen Werkzeugen und Technologien in sinnvollem Umfang arbeit innerhalb des Arbeitskreises ermöglicht es, Positionen und Zeitpunkt deterministisch erlauben, eigene Verhaltensweisen zu ley-Ideologie, wonach es keinen kategorialen Unterschied zwischen und an passender Stelle in Schule, Ausbildung, Studium und Weiter- Empfehlungen zu erarbeiten, die die vielschichtigen Aspekte der einem späteren Zeitpunkt zu beschränken. In sozialen Beziehun- menschlichen Individuen und Softwaresystemen gibt. Diese seit bildungen verankert sein müssen. Neben den Herausforderungen Digitalisierung berücksichtigen. gen könnten sich durch solch eine deterministische zukünftige John Searle als „starke künstliche Intelligenz“ bezeichnete Position in der Aus- und Weiterbildung steht aber auch die Softwaretechnik Selbstbindung gegenwärtig neue Handlungsmöglichkeiten er- verneint die besondere Verantwortung menschlicher Akteure und selbst vor neuen Herausforderungen. Unternehmen befinden sich Das vorliegende Papier gibt einen Überblick über ausgewählte Fra- öffnen. In paternalistischen Situationen hingegen dürften sie zu personalisiert digitale Systeme. Sie steht damit im Kontrast zu An- in einem kontinuierlichen Kulturwandel, der neue Kompetenzen gestellungen, die innerhalb des Arbeitskreises vertieft diskutiert einer deutlichen Bevorzugung der langfristigen Präferenzen einer sichten, die den Menschen die zentrale Rolle in der Digitalisierung erfordert in Hinblick auf Anforderungsprofile für Fachkräfte, Um- und untersucht werden sollen. Im Rahmen von Workshops, For- Person gegenüber den kurzfristigen Präferenzen mit ambivalenten zusprechen. Für die zukünftige Entwicklung der Digitalisierung und gang mit neuen Geschäftsmodellen, verkürzten Entwicklungs- und schungskooperationen und gemeinsamen Antragsinitiativen sollen Konsequenzen führen. Die Implikationen dieser Technologien und deren Auswirkungen auf die Gesellschaft ist es von hoher Bedeu- Innovationszyklen oder Verwertungs- und Schutzmöglichkeiten für zu diesen ausgewählten Fragestellungen eigene Positionspapiere Systeme auf die individuelle Freiheit in Entscheidungssituationen tung, welche Sichtweisen sich hier durchsetzen. Der Arbeitskreis neu entwickelte Produkte und Dienstleistungen in globalen Märk- erarbeitet werden, die politischen Handlungsbedarf konkretisieren sollen diskutiert und für konkrete Anwendungsbeispiele präzisiert möchte diesen Diskurs aktiv mitgestalten und die mit verschiede- ten. Hier gilt es in Zusammenarbeit mit dem Arbeitskreis Software und Handlungsempfehlungen darlegen. werden. nen Menschenbildern verbundenen Implikationen für die digitale Engineering Impulse zu setzen für eine verantwortungsbewusste Gesellschaft aufzeigen. und interdisziplinär ausgerichtete Gestaltung von digitalen Kompe- I.) Individuelle Selbstbestimmung und Freiheit in einer digita- Transparenz, Erklärbarkeit und Zurechenbarkeit beim Einsatz von tenzen und einen professionellen und offenen Umgang mit Soft- len Gesellschaft Algorithmen Akzeptanz automatisierter Entscheidungen: Da der Digitalisierung ware und digitalen Medien. Transparenz und Erklärbarkeit von Algorithmen: Vermehrt wird die in neuen Domänen häufig Skepsis entgegenschlägt, kommt der Selbstbestimmung, Datenschutz und Datenzugang Forderung vertreten, dass Algorithmen und deren Entscheidun- Akzeptanzforschung eine wesentliche Rolle zu. So kann beispiels- Gleichzeitig ist es notwendig, sicherzustellen, dass Menschen ihre Selbstbestimmung im Kontext Big Data: Zunehmende Vernetzung und gen transparent sein sollten. Die juristische Diskussion beschäf- weise im Bereich autonomer Systeme beobachtet werden, dass digitalen Kommunikations- und Arbeitsprozesse verstehen und Digitalisierung erlauben die Sammlung immer größerer Datenmen- tigt sich dabei vor allem mit Transparenzvorgaben für spezifische Menschen Vorbehalte gegen den Einsatz von Künstlicher Intelligenz anpassen können. Bei vielen analogen Arbeitsprozessen können gen. Insbesondere die Erhebung personenbezogener Daten mittels Anwendungsfälle, wohingegen die ethische Diskussion Trans- haben, insbesondere wenn diese moralisch relevante Entscheidun- Benutzer und Gruppen selbständig ihr Umfeld organisieren und Trackingtechnologien, die sich nicht mehr alleine auf Online-Um- parenz üblicherweise als normative Prämisse setzt. Im demokra- gen zu treffen hat. Hier ist wissenschaftliche Forschung notwendig, modifizieren. So kann ein Arbeiter die Anordnung seiner Werkzeu- gebungen beschränken, ermöglicht es, präzise und umfangreiche tischen Prozess ist dieser Top-Down-Ansatz problematisch, da es um die – gelegentlich diffusen – Bedenken der Bevölkerung und ge am Arbeitsplatz so anpassen, dass sie ergonomisch günstig und Profile über einzelne Personen zu erstellen. Im Spannungsfeld zwi- sich bei Transparenz aus Sicht der Betroffenen empirisch um ein unterschiedlicher Stakeholdergruppen zu verstehen und ihnen ge- effizient ist. Teams können interne Dokumenten-Workflows selbst schen neuen Innovationsmöglichkeiten durch Big Data und infor- zweischneidiges Schwert handelt. In anreizkompatiblen Experi- gebenenfalls mit einer geeigneten Kommunikation begegnen zu organisieren. Und Bürger können Angaben in einem Antragsformu- mationeller Selbstbestimmung ergeben sich sowohl Fragen nach menten wurde beispielsweise gezeigt, dass Menschen bestimmte können. lar einfach durch Kommentare am Seitenrand ergänzen. Digitale der technischen Ausgestaltung der eingesetzten Informationssyste- Informationen vermeiden um ihr Selbstbild zu wahren. Zudem ist Workflows bieten zwar auf der einen Seite neue Flexibilität. Die- me und Analysemethoden als auch nach den rechtlichen Rahmen- in Anwendungsfällen, in denen das Ergebnis algorithmischer Ent- Einfluss auf moralisches Verhalten: Die Digitalisierung beeinflusst se geht aber oft nur so weit, wie deren Entwickler sie implemen- bedingungen der Datenerhebung und -nutzung. Der Arbeitskreis scheidungen gerade nicht von den Betroffenen antizipiert werden wiederum das moralische Verhalten von Menschen. Daher soll tiert haben. Da digitale Systeme oft schwerer zu verändern sind, beschäftigt sich in diesem Kontext mit Datenschutztechniken, die soll, Transparenz kein relevantes Kriterium, weil die Offenlegung empirisch untersucht werden, inwieweit sich eine analoge Moral, als deren analoge Vorgänger, besteht die Gefahr, dass unflexible es Nutzern ermöglichen, sich vor ungewollter Datenerhebung tech- zu Manipulation einlädt. Ein Beispiel ist die Auswahl von Fällen zur die durch Face-to-Face-Interaktionen gekennzeichnet ist, von einer Software veraltete Workflows zementiert und zu einem Innova- nisch zu schützen oder eigene Daten zu anonymisieren. Für den Steuerprüfung, bei der die Betroffenen ein Anrecht auf die Rich- digitalen Moral unterscheidet. In diesem Zusammenhang gewinnt tionshindernis wird. Benutzer sollten deshalb in die Lage versetzt Arbeitskreis steht aber nicht allein die technische Machbarkeit im tigkeit und Rechtmäßigkeit der Entscheidung haben und zugleich die ergonomische Gestaltung digitaler Umgebungen eine ethische werden, Systeme kontinuierlich an ihre Bedürfnisse anzupassen.

42 | ARBEITSKREISE ARBEITSKREISE | 43 ARBEITSKREISE

Zum einen muss hierzu sowohl die Kompetenz aller Menschen ge- und Kunst-Präsentation. Es verändert sich auch das Selbstbild des stärkt werden, die sie umgebende digitale Welt zu modifizieren und Künstlers. Zur selben Zeit versuchen etablierte Kunstinstitutionen, zu erweitern. Hier müssen in der Bildungspolitik differenzierte Lö- ihre Tradition zu wahren und gegen eine als zu weitgehend emp- sungen evaluiert und vorangetrieben werden. Zum anderen muss fundene Auflösung zu immunisieren. auch sichergestellt werden, dass digitale Systeme konfigurierbare Benutzeroberflächen und offene Schnittstellen bieten. Hier können Digitalisierung und kulturelle Bildung: Welche Potenziale bietet die staatliche Institutionen und Entscheidungsträger sowohl regula- Digitalisierung im Bereich der kulturellen Bildung für Museen, torisch, als auch bei der Gestaltung von Ausschreibungen erheb- Archive und andere Kulturinstitutionen? Kulturelle Einrichtungen lichen Einfluss nehmen. nutzen zunehmend den digitalen Raum als Erweiterung. An vielen Orten und in unterschiedlichen Institutionen werden im Zuge von Institutionelle Transformationen Projekten Möglichkeiten entwickelt und erprobt, die Interaktion mit Digitale Transformationen sind fast immer auch institutionelle den Nutzern, Besuchern und Verbrauchern zu intensivieren, Häu- Transformationen: Nicht nur Unternehmen, sondern auch Be- ser zu öffnen, Barrieren abzubauen. Auf der anderen Seite sind ir- hörden, staatliche Stellen, öffentlich-rechtliche Einrichtungen, rationale Ängste weit verbreitet, eine zunehmend digitale Präsenta- Nicht-Regierungs-Organisationen, Kirchen oder Schulen sind mit tion und ein Zugang zu den Objekten via Internet würde langfristig neuen Herausforderungen beschäftigt. Diese beziehen sich nicht dazu führen, dass die Besucher die Kulturinstitutionen nicht mehr nur auf neue Kompetenzen und neue Erwartungen seitens der in Realität besuchen. Daneben stehen Fragen zum Verhältnis von Bürger, Mitglieder oder Schüler, sondern vor allem darauf, dass Original und digitaler Reproduktion beispielsweise in Form eines bestehende Regeln, Organisationsstrukturen und Öffentlichkeiten 3D-Drucks, und nicht zuletzt wird im Zuge des allgemeinen Kultur- unter Veränderungsdruck stehen. Digitale Infrastrukturen spielen wandels auch der Bildungsbegriff zu diskutieren und neu zu defi- eine zentrale Rolle beim Umbau und Aufbau nationaler, transna- nieren sein. tionaler und globaler Institutionen: zu denken ist etwa an zentrale europäische Informationssysteme wie Eurodac, SIS II oder Eurosur, Der Arbeitskreis setzt sich mit den formulierten Fragen auseinan- an offizielle und inoffizielle globale Finanzmärkte, wie sie z.B. in den der und identifiziert Maßnahmen, die geeignet sind, den kultur- globalen Bitcoin-Netzwerken entstehen oder an aktuelle Diskussio- technischen Wandel gemäß gesellschaftlicher Ziele und Werte zu nen über die Schaffung europäischer Plattformen für Forschungs-, gestalten. Medien- und Kunstförderung.

Auch die Digitalisierung der Wirtschaft erfordert erhebliche institutionelle Anpassungen, insbesondere in Form neuer Wettbewerbs- AUTOREN regeln. Die Geschwindigkeit des technischen Fortschritts und eine Daniel Schnurr (Universität Passau – ZD.B) hohe Unsicherheit stellen Gesetzgeber und Behörden dabei vor Dominik Herrmann (Universität Bamberg – ZD.B) besondere Herausforderungen. Da sowohl Untätigkeit als auch Daniel Méndez (TU München – ZD.B) falsche Entscheidungen in einem solchen Umfeld mit hohen so- Nicholas H. Müller, (FHWS – ZD.B) zialen Kosten verbunden sind, bedarf es der Vermittlung aktueller Jan-H. Passoth (TU München) theoretischer Erkenntnisse der Forschung in Form von umsetz- Matthias Uhl (TU München – ZD.B) baren Handlungsoptionen. Der Arbeitskreis will auf diese Weise Raphael Wimmer (Universität Regensburg – ZD.B) einen Beitrag zur Debatte um einen adäquaten Ordnungsrahmen für digitale Märkte leisten. Im Fokus stehen dabei der Umgang mit Marktmacht im Kontext datengetriebener Geschäftsmodelle und digitaler Plattformen sowie in der Praxis regelmäßig diskutierte Neutralitätsverpflichtungen (z.B. Netzneutralität, Datenneutralität).

IV.) Kulturtechnischer Wandel

Wandel der Kulturtechniken: Verschiedene Kulturtechniken verlieren durch die Digitalisierung zunehmend an gesellschaftlicher Anerkennung, was bis zum Verlust von Kompetenzen führen kann (z.B. Handschrift, Navigation, Auswendiglernen von Fakten). Auf der anderen Seite hat die Digitalisierung verschiedene neue Kul- turtechniken hervorgebracht (z.B. Bildbearbeitung, Query-Formu- lierung, kollaboratives Editing, Versionskontrolle). Wie lässt sich dieser Wandel in den Kulturtechniken beschreiben und welche Be- deutung muss ihm beigemessen werden? Welche Rolle spielen alte und neue Kulturtechniken für zentrale gesellschaftliche Einrichtun- gen (z.B. Bürokratie, Handel, Rechtsprechung)?

Digitalisierung in der Kunstpraxis: Gegenwärtig gibt es eine starke Tendenz, die verschiedenen Sparten der Kunst durch digitale Tech- niken zu vernetzen und traditionelle Grenzen zu überschreiten. Da- durch entstehen neue interdisziplinäre Formen in der Kunst-Praxis

44 | ARBEITSKREISE ARBEITSKREISE | 45 Lehre an den Hochschulen als auch der Prozessumsetzung im pri- ARBEITSKREISE vatwirtschaftlichen Bereich. Ein solcher partizipatorischer Design- prozess ist ebenfalls förderlich für die Erfolgsaussichten von Pro- dukten und Diensten am Markt.

Handlungsempfehlungen • Stärkung der Grundlagenforschung im Bereich der Mensch-Com- puter Interaktion, um ein tieferes Verständnis von Mensch und HUMAN-CENTERED ENGINEERING Mensch-Computer System zu erlangen. • Stärkung der Lehre im Bereich der Mensch-Computer Interaktion, POSITIONSPAPIER um die Kompetenzen im Design von Mensch-Computer Systemen in der bayerischen Wirtschaft und Gesellschaft zu verbessern.

"The best way to predict the future is to create it." Folge häufig das Leben vieler Menschen und ganze Gesellschaften "We shape our buildings; thereafter they shape us." Abraham Lincoln stark. Beispielsweise ist der Einfluss des Designs sozialer Medien Winston Churchill (Twitter, Facebook) auf unsere demokratischen Gesellschaftsstruk- Während in der Informatik traditionell die Automatisierung von turen heute unverkennbar. Deser Einfluss ergibt in großem Maße Arbeitsabläufen einen großen Stellenwert besitzt, entfalten viele aus den technischen Entwurfsentscheidungen, welche die Ent- Computersysteme erst in der Interaktion mit Menschen ihr wah- wickler getroffen haben. Neben nachträglichen regulierenden Ein- AUTOREN res Potential. Im Zuge der Industrialisierung war es vor allem die griffen durch den Gesetzgeber ist hier notwendig auch ein neues Albrecht Schmidt (LMU München – ZD.B) körperliche Arbeit der Menschen, die durch Werkzeuge und Ma- Designparadigma zu etablieren, welches den Menschen nicht nur Jörg Müller (Universität Bayreuth – ZD.B) schinen erleichtert und verstärkt wurde. Ähnlich ist es heute die als Benutzer und Kunden, sondern aus humanistischer Perspektive Florian Alt (Universität der Bundeswehr) geistige und intellektuelle Arbeit der Menschen, welcher wir durch eben zuerst und vor allem als Menschen begreift. Jens Grubert (HAW Coburg – ZD.B) interaktive Systeme nie vorher dagewesenes Potential verleihen. Andreas Muxel (Hochschule Augsburg – ZD.B) Insbesondere geben wir den Menschen durch Computersysteme Dieser Ansatz der Mensch-zentrierten Gestaltung ist das Ziel des Eva Rothgang (OTH Amberg-Weiden – ZD.B) sofortigen Zugriff auf fast das gesamte Wissen der Menschheit Arbeitskreises Human Centered Engineering. Dazu sollten eine Rei- und und enorme Datenmengen, aus denen sich neues Wissen ex- he von Prinzipien im Mittelpunkt des Designs stehen: trahieren lässt. Es stehen einer breiten Masse die digitalen Werk- • Der Mensch, seine Würde, Freiheit und Gesundheit, stehen im zeuge und die Rechenkraft zur Verfügung, um diese Datenmengen Mittelpunkt des Designs technischer Systeme. zu analysieren und zu nutzen. Instantane und multimediale Kom- • stehen und im gesamten Et Lebenszyklus von Produkten als . munikations- und Kollaborationsmöglichkeiten ermöglichen eine • Menschen sollten zu jedem Zeitpunkt die Kontrolle über techni- einfache und globale Zusammenarbeit für allen Menschen. Solche sche Systeme behalten und die Möglichkeit zur Intervention ha- Mensch-Computer Systeme besitzen enorme Problemlösefähigkei- ben. ten, um die Herausforderungen, die sich der Menschheit stellen, • Die Vorgehensweisen von Autonome Systeme und die Entschei- zu lösen. dungsprozesse sollen nachvollziehbar und transparent für den Menschen sein. Die Fähigkeit, interaktive Systeme mit optimaler Benutzbarkeit zu • Technische Systeme sollten Menschen vor allem die freie Entfal- gestalten, hat sich zur Kernkompetenz und zum marktentschei- tung ihrer Persönlichkeit im Kontext einer freiheitlichen Gesell- denden Merkmal vieler Unternehmen entwickelt. Systeme mit schaft ermöglichen. schlechter Benutzbarkeit sind im globalen Markt nicht mehr er- • Technische Systeme müssen unsere freiheitlich-demokratische folgreich. Dies haben insbesondere amerikanische Unternehmen, Grundordnung achten und sollten sie unterstützen. z.B. Apple, Google und Microsoft, früh erkannt, und sich so einen Entwicklungsvorsprung bei digitalen Produkten erarbeitet. Auf der Um diese Prinzipien zu achten, können wir spezifische Zielen for- anderen Seite haben bayerische Unternehmen in vielen Bereichen, mulieren. insbesondere Automotive, Automatisierung, Robotik und Medizin- technik, einen Vorsprung, den es zu erhalten und auszubauen gilt. Um den Menschen in den Mittelpunkt des Designs und der Ent- Hierbei spielt die Benutzbarkeit dieser Systeme eine zentrale Rolle. wicklung von technischen Systemen stellen zu können, ist ein tiefes So werden wir zukünftig unseren Arbeitsraum, Verkehrsraum und Verständnis des Menschen und der Mensch-Computer Interaktion Lebensraum mit antizipativen und autonomen Systemen teilen. notwendig. Insbesondere können Geschäftsmodelle, welche auf Entsprechende technische und rechtliche Rahmenbedingungen einer reinen Optimierung von Zielgrößen wie Klicks, Views und stellen immer noch eine große Herausforderung dar und werden Verweildauer beruhen, schädlich für Benutzer, Gesellschaft und aktuell kontrovers diskutiert. Es gilt hierbei das Vertrauen der Nut- Demokratie sein, wie jüngste Entwicklungen in sozialen Medien zer in neue Technologien zu stärken und sinnvolle Kommunikati- gezeigt haben. Statt die Benutzer zu befragen und fertige Produk- ons- und Interaktionsformen zwischen Mensch und Maschine zu te mit Nutzern zu testen, ist ein tiefes Verständnis der Benutzer, finden und zu etablieren. Folglich müssen wir zukünftige Szenarien ihrer Bedürfnisse und Probleme notwendig und dieses Verständnis schon heute erlebbar machen, um potenzielle Anwendungsgebiete muss im gesamten Entwicklungs- und Lebenszyklus des Produkts im jeweiligen soziokulturellen Kontext zu bewerten und mögliche berücksichtigt werden. Dies ist sowohl eine Frage der Grundlagen- Chancen, aber auch Risiken aufzuzeigen. forschung im öffentlichen Bereich als auch des Entwicklungspro- zesses im privatwirtschaftlichen Bereich. Eine zentrale Herausforderung ist, dass wir durch Designent- scheidungen für die Entwicklung digitaler Systeme, und hierbei Um den Menschen ihre Entfaltung in Arbeit und Leben zu ermög- insbesondere Entscheidungen für das Interaktionsdesign und die lichen, ist eine Beteiligung der von einem System betroffenen Schnittstelle Mensch-Maschine, in der Lage sind, Menschen (Indi- Personen in einem partizipatorischen Designprozess dringend zu viduen, Gruppen und ganze Gesellschaften) neue perzeptuelle und empfehlen. Insbesondere dürfen Menschen nicht nur als Benutzer kognitive Fähigkeiten zu verleihen. betrachtet werden, sondern sollten als freie Individuen Einfluss auf die von Ihnen benutzen Systeme und Ihre Daten haben. Dies kann Heute werden zentrale Designentscheidungen oft innerhalb von insbesondere durch eine Teilhabe nicht nur im Designprozess, son- Privatunternehmen aus technischer und wirtschaftlicher Perspekti- dern auch im Betriebs-, Wartungs- und Weiterentwicklungsprozess ve getroffen. Diese Designentscheidungen beeinflussen aber in der von Systemen sichergestellt werden. Dies ist sowohl eine Frage der

46 | ARBEITSKREISE ARBEITSKREISE | 47 • Technische Kompetenzen als Nicht-Verrechenbarkeit skizziert wird. Wir müssen deshalb Ant- ARBEITSKREISE • Methodenkompetenzen und weitere überfachliche Kompetenzen worten auf die Frage finden, wie in der Ausbildung des Software (wie Software Engineering-spezifische Soft Skills), etwa wenn es Engineering ein digitaler Humanismus die menschliche Verantwor- um den Umgang mit Anforderungen und damit verbundener Er- tungsfähigkeit stärken kann. hebungs- oder Verhandlungstechniken oder um praktische Pro- jektorganisation und Management geht.

Gleichzeitig ist die Anzahl der Studierenden und der Bedarf an AUTOREN SOFTWARE ENGINEERING qualifiziertem Software Engineering Nachwuchs auf einem Rekord- Daniel Méndez (TU München – ZD.B) hoch. Jürgen Mottok (OTH Regensburg – ZD.B) POSITIONSPAPIER Jörg Abke (HAW Aschaffenburg) Eine Bedarfserhebung und die damit verbundene Schaffung neuer Gerd Beneken (Hochschule Rosenheim) Stellen (Mittelbau und Professuren) in der bayerischen Hochschul- Axel Böttcher (Hochschule München) Während Software vor wenigen Jahren noch eine eher untergeord- Neben der erforderlichen Reflexion über die notwendigen -Inhal landschaft stellt dabei einen wesentlichen und notwendigen Mei- Konrad Doll (HAW Aschaffenburg – ZD.B) nete Rolle in einer von Hardware dominierten Welt einnahm, konnte und Formate innovativer Software Engineering Themen in der lenstein dar. Die Förderung von Projekten, die sich vor allem der Georg Hagel (HAW Kempten) te sich das Software Engineering (dt.: Softwaretechnik) mittlerwei- Lehre gilt es, die Zugänge zu den Basiskompetenzen der Software- systematischen Verbesserung der Lehre widmen, sollte als weitere Dieter Landes (HAW Coburg) le u.a. dank fortschreitender Standardisierung von Hardware als technik für andere Disziplinen, etwa im ingenieur-technischen oder Möglichkeit in Betracht gezogen werden. Diese Projekte wären im Jan-H. Passoth (TU München) eigenständige Disziplin emanzipieren und treibt seither in ganzen wirtschaftlichen Bereich, zu öffnen. Softwaretechnik ist insgesamt Bereich (fach)didaktischer Fundierung oder der Erprobung neuer, Alexander Popp (Universität der Bundeswehr München) Industrien Innovationen und Transformationsprozesse an. Soft- nicht mehr allein in der Hochschullehre der Informatik essentiell, auch interdisziplinärer, Lehrformate angesiedelt. Sie böten, z.B. als Yvonne Sedelmaier (HAW Coburg) ware nimmt heute eine zentrale Rolle bei der digitalen Transforma- sondern immer mehr Disziplinen benötigen Grundkenntnisse im Verbundprojekte aufgesetzt, gleichzeitig eine hervorragende Gele- Janet Siegmund (Universität Passau – ZD.B) tion ein. Die Softwaretechnik spielt dabei aufgrund der Dominanz Software Engineering, noch dazu in wachsendem Umfang. Daher genheit, Synergien zwischen Universitäten, an denen eher grundla- der Software in neuen Technologien eine besondere Rolle: Soft- müssen wir ein gemeinsames Verständnis für ein Basiscurriculum genorientierte Themen verankert sind, und Hochschulen für ange- ware wird immer komplexer und ist allgegenwärtig. Vor allem aber erarbeiten, das die Vielfalt an Themen und Methoden auch außer- wandte Wissenschaften, an denen eher anwendungsnahe Themen durchdringen heute software-intensive Technologien sämtliche halb der Informatik transportiert, ohne dabei in Beliebigkeit zu ver- untersucht werden, gezielt zu nutzen. Lebensbereiche moderner Gesellschaften. Deshalb müssen einer- fallen. Gleichzeitig kann Software Engineering und seine Lehre an seits junge Menschen besser auf die Nutzung der Potenziale soft- Hochschulen von methodischen Zugängen in andere Disziplinen Stärkung der Forschung und Entwicklung ware-intensiver Technologien, deren Risiken und die Anforderun- stark profitieren, etwa den Geistes- oder Sozialwissenschaften, um Die Forschung muss auf allen Ebenen ein stärkeres Bekenntnis zu gen der Wissensgesellschaft vorbereitet werden; Bildung für einen den notwendigen Brückenschlag von einer bislang gestaltungsori- Software Engineering als erkenntnisorientierte und evidenzgetrie- sicheren, ressourcenschonenden und verantwortungsbewussten entierten in eine erkenntnisorientierte Disziplin zu schaffen. Soft- bene Disziplin erfahren. Ohne fundierte empirische Forschung wird Umgang mit Softwarelösungen, etwa im Umfeld von IKT-Anwen- ware wird von Menschen für Menschen erstellt; die damit verbun- es nicht gelingen, die Grenzen neuer Methoden und Technologien dungen, und der Ausgestaltung des sozialen Miteinanders kommt denen Prozesse unterliegen daher auch kulturellen Normen und reproduzierbar auszutesten und den Transfer grundlagenorientier- in einer digitalisierten Welt wachsende Bedeutung zu [1]. Anderer- Gesetzen, die es grundlegend zu erforschen und zu durchdringen ter Beiträge in die Praxis gezielt zu unterstützen. Dieser Wandel er- seits liegt eine mindestens ebenso große Herausforderung darin, gilt. Ebenso spielt der Mensch im Software Engineering eine zen- fordert insbesondere: künftige Software-Ingenieure auszubilden. Insbesondere die Ent- trale Rolle – der Erfolg von Softwareentwicklungsprojekten hängt • Schaffung von Möglichkeiten zur Förderung von Themen mitex- wicklung komplexer Softwaresysteme erfordert neben fachlichen zum einen stark von einem funktionierenden Entwicklungsteam perimentellem Charakter an Hochschulen, ohne dabei zwingend in Kompetenzen ein hohes Maß an interdisziplinären und überfach- und zum anderen vom Funktionieren disziplinübergreifender Kom- ein Konsortium mit Industrievertretern treten zu müssen, gleich- lichen Kompetenzen. Hier muss das Bildungssystem in die Lage munikationsprozesse ab. zeitig aber eine hohe Anwendungsrelevanz der Themen sicherzu- versetzt werden, kontinuierlich auf hohem Niveau Software-Inge- stellen. nieure auszubilden. Damit der Brückenschlag hin zu einem vermehrt evidenzgetriebe- • Schaffung von Möglichkeiten zur Förderung von Projekten in internen Software Engineering gelingt, muss das empirische Software disziplinären Konfigurationen und (Forschungs-)Methodenvielfalt. Software Engineering treibt ohne Frage den digitalen Wandel wie Engineering mehr Wertschätzung in der Forschung erfahren. Dar- [3] Die Kombination von beispielsweise Psychologie, Sozialwissen- keine andere Disziplin voran und befindet sich selbst in einem Wan- über hinaus müssen bessere Zugänge zu Fördermöglichkeiten für schaften und Software Engineering erlaubt es Fragestellungen zu del. Diesen Wandel zu durchdringen und konkrete Vorschläge für Forschungsprojekte an Hochschulen mit hohem experimentellen beantworten, die einzelne Disziplinen nicht umfassend adressieren die Gestaltung von Software Engineering Themen zu erarbeiten ist Charakter geschaffen werden. Der Wandel in eine erkenntnis- können, etwa die Fragen, wie Mechanismen der Selbstorganisation das Ziel des ZD.B Arbeitskreises Software Engineering. orientierte Disziplin ist in Forschung und Praxis bereits in vollem funktionieren und wie diese verbessert werden können. Gange, sollte sich aber auch etwa in öffentlichen Förderlinien wie- Insgesamt stellt der digitale Wandel die Wirtschaft, die Forschung derspiegeln, damit sich Forschung auch losgelöst von konkreten in- Stärkung ethisch reflektierten Handelns und die Lehre gleichermaßen vor neue Herausforderungen. Zum dustrierelevanten Fragestellungen und Problemen entfalten kann. Die Gestaltung von Informations- und Kommunikationstechnologie einen herrscht in Bayern ein massiver Bedarf an qualifizierten Mit- Um Bayern nachhaltig im Zuge der Digitalen Transformation und erfordert nicht nur die Berücksichtigung technischer Leistungsan- arbeiterinnen und Mitarbeitern, der aktuell nicht gedeckt werden dem damit verbundenen Wandel im Software Engineering zu stär- forderungen, sondern auch die Einbeziehung ethischer Überlegun- kann. In 2017 gab es 55.000 offene Stellen mit Informatikbezug in ken, sehen wir insbesondere in folgenden Bereichen unmittelbaren gen. Die voranschreitende Digitalisierung hat eine tiefgreifende Deutschland, davon 63% in der Software-Entwicklung.[2] Zum an- Handlungsbedarf, dem wir uns um Rahmen des Arbeitskreises Veränderung der Gesellschaft und aller ihrer Lebensbereiche zur deren müssen Inhalte und Formate - also insgesamt die Didaktik widmen möchten. Folge. Um sowohl am menschlichen als auch am ökologischen des Software Engineering - (weiter)entwickelt und überdacht wer- Wohlergehen orientierte ökonomisch wie technisch verantwor- den, um dem Anspruch Software Engineering als Schlüsselkompe- Stärkung des Lernens und der Lehre tungsbewusste Handlungs- und Entscheidungskriterien entwickeln tenz zu stärken und dem kontinuierlichen Wandel des Software Aktuell könnte oft der Eindruck entstehen, dass bei der Verbreite- zu können, wird eine kritische Reflexion von Chancen und Risiken Engineering gerecht zu werden. Zum Beispiel, welchen Stellenwert rung der Kompetenzen an staatlichen bayerischen Hochschulen bereits jetzt gängiger Praktiken und zukünftiger Möglichkeiten di- haben neue Vorgehensmodelle, etwa agile Methoden, in der Wirt- und der damit verbundenen Schaffung neuer Professuren, etwa gitaler Technologien benötigt [4]. Kann ein rationales (autonomes) schaft? Welchen in der Lehre? Wie lassen sich Kompetenzen im im Bereich Data Science and Engineering oder Artificial Intelligence, IT-basiertes System mit softwarebasierten Algorithmen so gesteu- Software Engineering zielgerichtet adressieren? Wie kann Metho- die kontinuierliche und nachhaltige Stärkung grundlegender Kom- ert werden, dass seine Entscheidungen den Erwartungswert der denwissen, etwa zu agilen Methoden, praxisnah und praxisrelevant petenzen im Software Engineering zwar als wichtig erachtet, diese Handlungskonsequenzen optimieren kann? Die utilitaristische eingeübt werden? aber dennoch vernachlässigt und geschwächt wurden. Solche Ba- Ethik, die als Maxime die nützliche Machbarkeit von IT-Technologie siskompetenzen umfassen: ansieht, kollidiert unter anderem mit einem fundamentalen Prinzip jeder zivilen und humanen Gesellschaft, das von Nida-Rümelin [5]

[1] Siehe auch Hightech-Strategie Innovationen für Deutschland, Bundesministerium [3] Siehe auch “Empirical Software Engineering: From Discipline to Interdiscipline, für Bildung und Forschung https://arxiv.org/abs/1805.08302 [2] Bitkom Studie aus 2017 https://www.bitkom.org/Presse/Anhaen- [4] Siehe “Das Ethik-Manifest des IT-Sicherheitscluster Bayern e.V.” ge-an-PIs/2017/11-November/Bitkom-Charts-IT-Fachkraefte-07-11-2017-final.pdf [5] Siehe Julian Nida-Rümelin: Digitaler Humanismus, Die Zeit, 1.10.2017

48 | ARBEITSKREISE ARBEITSKREISE | 49 onsprozessen und eine Explosion an Sensorik werden zunehmend einer Vorlaufzeit zu rechnen ist bis die ersten Absolventen die Aus- ARBEITSKREISE große Datenmengen erzeugt, die eine Grundlage bilden für KI-An- bildung durchlaufen haben. wendungen wie Condition-Monitoring und Anomalie-Detektion, Auslastungsprognose und Trenddetektion, Ursachenfindung bei Um die Zwischenzeit zu nutzen, wollen wir die Zusammenarbeit Störungen und Meldungsgewittern, Digitale Zwillinge, KI-basierte zwischen Unternehmen und Experten auf dem Gebiet der KI durch Prozessmodellierung und Verbesserung in Regelung und Steuerung, Förderung von Co-Innovation unterstützen. Hackathon-Events bie- und Robotic Process Automation. In der Energieerzeugung kann die ten die Möglichkeit zur interdisziplinären Zusammenarbeit zwischen Umweltbelastung signifikant reduziert werden, beispielsweise durch Unternehmen und Studierenden. Einzelne von Unternehmen ausge- ARTIFICIAL INTELLIGENCE eine Reduzierung des Ausstoßes giftiger Stickoxide von Gasturbinen tragene Events haben in der Vergangenheit schon Erfolg dabei ge- über selbstoptimierende Verfahren, ohne dabei die Leistung der habt, Experten aus der Privatwirtschaft und der Forschung für sich POSITIONSPAPIER Turbine zu schmälern oder ihre Lebensdauer zu verkürzen. KI wird zu gewinnen. Zum Beispiel veranstaltete Audi 2016 den “Smart Fac- eingesetzt, um den Betrieb von Windturbinen zu verbessern, die die tory Hackathon”, bei welchem 17 Teams neue Ideen und Prototypen Stellung der Rotoren autonom so dem wechselnden Wind anpassen, unter dem Motto “Smart Factory” entwickelten, in dem sie tatsäch- Die künstliche Intelligenz (KI) wird die Kerntechnologie des 21. Jahr- schritten. Mit Blick auf eine alternde Gesellschaft und aufwendige dass der Ertrag eines Windparks steigt. Eine intelligente Überwa- liche/echte/reale Unternehmensdaten benutzten. Die Ressourcen- hunderts werden. Die damit verbundenen wirtschaftlichen Konse- Behandlungsmöglichkeiten werden diese Kosten weiter steigen. KI chung und vorausschauende Wartung intelligenter Stromnetze kann intensität solcher Events macht deren Austragung durch einzelne quenzen und gesellschaftlichen Umwälzungen sind vergleichbar bietet die Möglichkeit Krankheiten bereits im Frühstadium zu er- Netzverfügbarkeit wesentlich verbessern. In der Bahntechnik kann Unternehmen aber kaum attraktiv. Aus diesem Grund schlagen wir mit denen der Elektrifizierung des Kontinents zu Beginn des ver- kennen und dadurch effektiver und kostengünstiger zu behandeln. über eine KI-gestützte Auswertung der Betriebsdaten der Fahrzeu- die Förderung von KI-Hackathon-Events vor, in welchen mehrere gangenen Jahrhunderts. Die KI ist der Motor für die Entwicklung Weiterhin unterstützt KI die effektive Verknüpfung von klinischen ge die Wartung intelligent geplant werden und es können so Aus- Unternehmen gleichzeitig Herausforderungen stellen, für welche sie zukünftiger intelligenter Produkte in vielen Branchen. Im KI-Be- Daten und Daten aus mobilen Systemen um damit geeignete Prä- fallzeiten reduziert werden. Zudem kann die KI helfen, die Leitzen- sich einen Beitrag erhoffen. Eine solche Bündelung von Ressourcen reich sind derzeit vor allem US-amerikanische Unternehmen wie ventionsansätze für individuelle Krankheitsrisiken zu bestimmen, trale eines Stellwerks optimal zu konfigurieren. Die Software wählt wird die Zusammenarbeit zwischen KI-Experten von Hochschulen Google, Facebook und Amazon führend und investieren Milliarden, sowie Patienten und Gesunde in Genesung und Gesundheitserhal- aus Milliarden Möglichkeiten für die Hardwarekonfiguration eines und Unternehmen verbessern. Indem KI-Hackathon-Events Fuß fas- wobei auch China durch ein umfangreiches Investitionsprogramm tung zu helfen.[2] Darüber hinaus hat KI das Potenzial personali- Stellwerks jene aus, die alle Rahmenbedingungen erfüllen – auch in sen und an Zuspruch gewinnen, wird die Position Bayerns als zentra- rasant aufholt. Im Vergleich dazu sind bayerische Unternehmen sierte Diagnose- und Therapieverfahren zu unterstützen, d.h. die puncto Betriebssicherheit. le Anlaufstelle/Hub in Fragen der KI gefestigt. eher zögerlich im Einsatz von KI-Technologien, obwohl großes richtige Therapie für den richtigen Patienten zur richtigen Zeit be- Marktpotenzial besteht. Falls Bayern den Anschluss im Bereich der reitzustellen. Aktuell entstehen hohe Kosten und Lebenseinschrän- Handlungsempfehlungen KI-Forschung verliert, leidet zum einen die Wettbewerbsfähigkeit kungen bei den Patienten durch die Anwendung fehlgeschlagener Mit Blick auf die vor uns liegenden politischen und technischen Ent- und zum anderen steigt die Abhängigkeit von ausländischen KI-Lö- Therapien. Für die Entwicklung von KI-Lösungen müssen zum einen wicklungen ist es von größter Wichtigkeit in Bayern Systemintegra- AUTOREN sungen. Gerade in der aktuellen Diskussion um ethische Fragestel- große Datenbestände aufgebaut und zum anderen intelligente Al- tionskompetenz für KI aufzubauen und diese möglichst rasch pro- Christian Wachinger (LMU München, Klinikum – ZD.B) lungen in der KI (repräsentiert die KI unsere Werte?) ist deshalb die gorithmen entwickelt werden, die auf Basis dieser Daten lernen duktiv in der Wirtschaft zu nutzen. Nur damit kann Bayern von den Alois Knoll (TU München, Fakultät für Informatik) Entwicklung eigener KI-Lösungen notwendig. Krankheitsentwicklungen vorherzusagen und Therapiemöglichkei- zu erwartenden globalen Entwicklungen profitieren. Zusätzlich muss Volker Tresp (LMU München und Siemens AG) ten anzubieten.[3] es uns jetzt darum gehen, die nächsten Schritte der KI-Forschung ak- Jens Grubert (HAW Coburg – ZD.B) Die Entwicklung im KI-Bereich steht noch am Anfang und durch In- tiv mitzugestalten. Auch die Grundlagenforschung kann von den in Konrad Doll (HAW Aschaffenburg – ZD.B) vestitionen in KI könnte Bayern rasch aufholen. Das gut funktionie- Der Bereich Automotive hat eine immense Stellung für den Aus- Bayern prävalierenden Themen (Gesundsheitswesen, Automotive, Jonas Biehler (TU München und AdCo Engineering GW GmbH) rende Ökosystem aus forschungsstarken Universitäten, großen bau und die Sicherung des Wirtschaftsstandorts Bayern. Moderne Industrie) stark profitieren wenn diese von solchen Anwendungen Patrick van der Smagt (Volkswagen Aktiengesellschaft) Technologieunternehmen und Konzernen sowie eine aktive Start- Fahrerassistenzssysteme und automatisiertes Fahren sind neben inspiriert wird, und entsprechende KI-Forschungsthemen verfolgt Oliver Amft (FAU, Lehrstuhl für Digital Health – ZD.B) Up-Szene in Bayern bieten nahezu ideale Voraussetzungen, um die Elektromobilität Kernthemen für die Zukunftsfähigkeit. Auch baye- werden. Luis I. Lopera González (FAU, Lehrstuhl für Digital Health – ZD.B) Entwicklung und Anwendung von KI voranzutreiben, sofern die nö- rische Automobilhersteller und -zulieferer arbeiten mit Nachdruck Andreas Muxel (HAW Augsburg – ZD.B) tigen Anschubfinanzierungen erfolgen. Frankreich hat zuletzt mit an diesen Themen. Audi hat mit dem „Stau-Pilot“ die erste serien- Wir begrüßen die Zukunftsinitiative für Künstliche Intelligenz mit den angekündigten Investitionen in Höhe von 1,5 Milliarden Euro reife automatisierte Fahrfunktion auf Autobahnen entwickelt und dem geplanten Kompetenznetzwerk „Künstliche maschinelle In- im Bereich der KI für Aufsehen gesorgt. Auch die EU hat kürzlich leitet in Bayern für den Volkswagen Konzern die Entwicklungen telligenz“ sowie die bayernweite Einrichtung von Kompetenz- und Pläne vorgestellt 20 Milliarden Euro in KI-Forschung zu investieren. auf dem Gebiet des vollautomatisierten Fahrens. Weitere Funktio- Forschungszentren. Um die Anstrengungen im Bereich der KI in Zudem begrüßen wir das von Ministerpräsident Söder angekündig- nen werden auch von anderen Herstellern und Zulieferern folgen. Bayern darüber hinaus zu bündeln, schlagen wir die Gründung des te Kompetenznetzwerk „Künstliche maschinelle Intelligenz“. Zentraler Bestandteil dieser Funktionen sind Methoden der KI und „Bayerischen KI-Zentrums (BaKIZ)“ vor. Am Zentrum sollte sowohl insbesondere des Maschinellen Lernens (ML). Als Beispiele sind international erklassige Grundlagenforschung im Bereich der KI Die großen US-amerikanischen Firmen entwickeln KI-Verfahren, die die Umfeldperzeption oder die Verhaltensprädiktion von Verkehrs- stattfinden, als auch die Erforschung von Anwendungen von KI in besonders gut für die Mustererkennung in großen Datenmengen teilnehmern bei der Situationsanalyse zu nennen. Der Forschungs- den Bereichen Medizin, Automotive, Automation und Robotik. Das und für Anwendungen in Massenmärkten geeignet sind, etwa zur und Entwicklungsbedarf ist erheblich. Einige Aspekte seien hier Zentrum soll außerdem in Zusammenarbeit mit dem Leibniz-Re- Vorhersage von Käuferverhalten aus Ruckmeldungen von Kunden exemplarisch angeführt: Der urbane Straßenverkehr stellt bzgl. der chenzentrum den Zugang zu einer skalierbaren, hochmodernen oder zur präzisen Zielgruppenansprache von Werbekunden.[1] Im Automatisierung wegen der vielfältigen und komplexen Verkehrs- und funktionsspezifischen KI-Infrastruktur ermöglichen. Gegensatz hierzu ist der KI-Bedarf in Bayern vor allem in den Be- situationen und der zusätzlichen Verkehrsteilnehmer, wie Fußgän- reichen der Medizin sowie insbesondere auch in den industriellen ger und Radfahrer, eine besondere technische Herausforderung Um den Wettbewerb im akademischen Bereich und Forschung Anwendungen – bspw. in der Robotik, der Automatisierung oder im dar. Außerdem müssen methodisch im Bereich der Sicherheit von von höchster Qualität zu fördern, schlagen wir vor, einen jährli- Automotive-Umfeld – anzusiedeln. KI-Anwendungen in diesen Be- Systemen, die auf ML beruhen, Lösungen erarbeitet werden. Auch chen Preis ins Leben zu rufen, der die beste KI-Arbeit prämiert. Ein reichen unterscheiden sich von KI-Anwendungen für den Massen- das effiziente Sammeln und Auswerten von Daten sowie die Inte- solcher Preis wird den Wiedererkennungswert der teilnehmenden markt: zum einen hinsichtlich der beschränkten Verfügbarkeit von gration in die Modelle sind umfangreiche Forschungsthemen. Auf Forscher steigern und gleichzeitig die hohe Qualität der Arbeiten annotierten Datensätze und zum anderen in der teils gravierenden der anderen Seite sind die Chancen, tragfähige und zukunftsfähige sicherstellen. Tragweite der von der KI zu treffenden Entscheidungen. Vor die- Lösungen zu erarbeiten, enorm. sem Hintergrund ist es sinnvoll, in Bayern vor allem fokussiert in Außerhalb des akademischen Umfelds ist die Anwendung von KI Medizin, Automotive, Automation und Robotik zu investieren, Für die industrielle Anwendung bietet die KI enormes Potenzial. zur Datenanalyse sehr verbreitet. Aus diesem Grund empfehlen um Bayern mittelfristig den entscheidenden Vorsprung in diesen Hier sind Investitionen auf breiter Ebene notwendig, um die gute wir, dass Universitäten und Hochschulen einen grundlegenden Technologien zu verschaffen und diesen auch langfristig zu sichern. Marktposition der deutschen Industrie zu stärken und auszubauen. Kurs für maschinelles Lernen in allen relevanten Studiengänge Wichtige Anwendungsgebiete sind Fertigung und Produktion, aber einbinden. Ziel ist es, die Bildung auf dem Gebiet der KI zu fördern Die Ausgaben im Gesundheitswesen in Deutschland haben im Jahr auch Mobilität, Energieerzeugung, Energieverteilung und smarte und die frühe Anwendung von KI zur Problemlösung und Daten- 2017 zum ersten Mal die Marke von 1 Milliarde Euro pro Tag über- Antriebe. Durch die Digitalisierung von Fertigungs- und Produkti- analyse zu fördern. Die Umsetzung sollte rasch geschehen, da mit

[1] Ausführlich Diskussion in: A. Knoll, Kunstliche Intelligenz und Robotik: Motor fur [3] Überblick in: Tresp, V., Overhage, J. M., Bundschus, M., Rabizadeh, S., Fasching, P. A., Innovation, ZD.B Digital Dialogue, 2018. & Yu, S. (2016). Going digital: A survey on digitalization and large-scale data analytics [2] Überblick in: Amft, O. (2018). How Wearable Computing Is Shaping Digital Health. in healthcare. Proceedings of the IEEE, 104(11), 2180-2206. IEEE Pervasive Computing, 17(1), 92–98.

50 | ARBEITSKREISE ARBEITSKREISE | 51 werden, [6] [7] sofern die technische Sicherheit der Systeme ge- Fall, wenn ein System nur teilautonom agiert, der Mensch jedoch ARBEITSKREISE währleistet ist. Gemäß der Weltgesundheitsorganisation sind denkt, es wäre vollautonom und so das System nicht genau über- Autounfälle die Haupttodesursache in der Altersgruppe der 15- wacht, oder der Mensch überschätzt die Möglichkeiten des teil- bis 29-jährigen. Durch den Umstieg auf autonome Fahrzeuge autonomen Systems. könnten Verkehrstote in einer Größenordnung von 1,25 Millio- 6. [Ges_R_Rebound_Effekt] Die mit dem autonomen Fahren ver- nen Menschen pro Jahr verhindert werden. bundenen Effizienzgewinne können durch den Rebound-Effekt 2. [Ges_C_Mobilität] Das autonome Fahren wird das Mobilitätsver- vernichtet werden. Rebound-Effekte bezeichnen Situationen, in ständnis der Gesellschaft verändern. Es ermöglicht Carsharing denen technisch erzielte Effizienzsteigerungen durch extensive AUTONOMOUS SYSTEMS in einer neuen Dimension, die den privaten Besitz unattraktiver Nutzung aufgezehrt werden [9]. Im Falle des autonomen Fahrens werden lässt. Die Hoffnungen der ökologischen Bewegung sehen könnte sich der Rebound-Effekt durch Zunahme des Verkehrs- POSITIONSPAPIER eine deutliche Verringerung der Fahrzeuge auf den Straßen, weil aufkommens oder durch Kannibalisierung des öffentlichen Per- der Besitz von eigenen autonom fahrenden Fahrzeugen durch sonennahverkehrs äußern. direkte Zugriffsmöglichkeiten nicht mehr so attraktiv sein könn- Autonome Systeme werden künftig in zunehmendem Maß den vernetzter Mobilitätsanforderungen einer sich ändernden Gesell- te: Mobilität würde unabhängig von Besitz möglich. Dadurch Handlungsempfehlungen menschlichen Alltag gestalten und für ausgewählte Teilbereiche schaft. kann die Anzahl der Fahrzeuge reduziert, die Umwelt entlastet In diesem Abschnitt werden die Risiken und Chancen des vorange- sogar die Kontrolle vollständig vom Menschen übernehmen. In den und ein neues Stadtbild entwickelt werden, das den Menschen gangenen Abschnitts in allgemeine Handlungsempfehlungen über- hochtechnisierten Gesellschaften stehen wir jetzt an der Schwel- Technologische und gesellschaftliche Herausforderungen in den Mittelpunkt rückt. Diese neue Art der Mobilität steht ins- führt. Die Handlungsempfehlungen verweisen durch die Referenz le zu einem Zeitalter, in dem uns Autonomisierung viel Sicherheit, Durch das Zusammenspiel verschiedener Sensoren, künstlicher In- besondere auch älteren Menschen und Menschen mit Behinde- auf die jeweilige Chance/das jeweilige Risiko, die/das diesen jeweils Komfort und Freiräume einräumen könnte. Damit die mit der telligenz und verteilter Softwaresysteme können Automobile schon rung zur Verfügung. Die Reisezeit im Fahrzeug kann produktiv zugrunde liegt (beispielsweise [Tec_R_Sicherheit]). Autonomisierung verbundenen Vorteile möglichst allen Menschen heute (teilweise) autonom fahren. Damit unterliegt das Automobil genutzt werden, sodass das Auto zu einem weiteren Lebensmit- sowie künftigen Generationen zugutekommen können, sollte der einer Wandlung, die stellvertretend und vorausweisend für den telpunkt wird. Die neue Nutzungsmöglichkeit der Reisezeit kann Lehren und Lernen anstehende Übergang reflektiert genutzt und proaktiv gestaltet Transformationsprozess durch die Digitalisierung steht. Nicht zu- die Anzahl an Pendlern und die Attraktivität ländlicher Gebiete • Verpflichtendes Kernelement jeder IT-Ausbildung ist die Vermitt- werden. letzt stellt die fehlende Akzeptanz innerhalb der Bevölkerung eine als Wohnraum steigern. lung umfassender Kenntnisse zu Verlässlichkeit von softwarein- Hürde dar, die es zu überwinden gilt. Umfragen zufolge ist nur gut 3. [Ges_C_Staatliche_Steuerung] Die mit der Autonutzung einherge- tensiven Systemen. [Tec_R_Sicherheit] In diesem Positionspapier wollen wir Handlungsempfehlungen ab- jeder Vierte bereit, sich von einem Computer chauffieren zu lassen. henden Externalitäten können durch den Einsatz dynamischer • Ethische Argumentation sowie empirische Erhebungen zu der Be- leiten, indem wir uns zunächst umfassend den Chancen und Risi- In diesem Abschnitt sollen die Chancen und Risiken des autonomen Mautsysteme und Staugebühren eingepreist werden. Durch die reitschaft in der Gesellschaft, Risiken einzugehen, müssen in der ken von autonomen Systemen widmen. Dabei soll das autonome Fahrens beleuchtet werden, die im darauf folgenden Abschnitt in gezielte Setzung von Steuern und Gebühren kann eine Feinab- Lehre thematisiert werden. Innovative Systeme der Zukunft agieren Fahren als ein exemplarischer Vertreter von autonomen Systemen Handlungsempfehlungen überführt werden. stimmung urbaner Transportsysteme erfolgen. So können Stadt- zunehmend an der Schnittstelle zwischen dem technisch Machbaren dienen, der aufgrund seiner Vorreiterrolle als Beispielanwendung planer Fahrten in ärmeren Gegenden subventionieren und im und dem gesellschaftlich Tragfähigen. Die Gesellschaft muss im Um- besonders geeignet ist. Unter autonomem Fahren verstehen wir Technologische Chancen und Risiken Falle längerer Autofahrten den öffentlichen Verkehr fördern. gang mit autonomen Systemen geschult werden. [Ges_R_Ethik, Ges_R_ die (Teil-)Automatisierung von Kraftfahrzeugen mit einem poten- Chancen Automatisierung, Ges_R_Teilautomatisierung, Ges_R_Rebound_Effekt] tiell möglichen Eingriff durch den Fahrzeuglenker. Die Eignung die- 1. [Tec_C_Errungenschaft][2] Errungenschaften in der Forschung an Risiken ser Beispielanwendung ergibt sich auch aus der Tatsache, dass die autonomen Automobilen werden sich positiv auf andere Berei- 1. [Ges_R_Fehlertoleranz_Akzeptanz] Der Mensch hat häufig Angst Forschung und Entwicklung Gesellschaft bereits vergleichsweise umfassend informiert ist be- che auswirken. Die Forschungsbereiche umfassen dabei Hard- vor einem autonomen System und traut ihm nicht. Das Risiko • Die Kenntnisse und Forschungsergebnisse zum Schutz und zur ziehungsweise sich aktiv in die Debatte einbringt, insbesondere bei ware, Software sowie Konzepte zum Datenschutz.[3] Innovatio- [Tec_R_Sicherheit] hat großen Einfluss auf die Akzeptanz in der Sicherheit softwareintensiver Systeme müssen sich schneller wei- der Untersuchung von Grenzfällen des autonomen Fahrens, bei- nen auf diesen Feldern werden Bereichen wie Smart Home und Gesellschaft. Ein Fehler im System kann die Reputation des auto- terentwickeln als die Angriffe auf diese. [Tec_R_Sicherheit, Ges_R_ spielsweise anhand von ethischen Dilemmata. Dies ist für die frü- Robotik zugutekommen und die Digitalisierung in weiteren Be- nomen Fahrens deutlich verschlechtern. Außerdem toleriert der Automatisierung, Ges_R_Fehlertoleranz_Akzeptanz] he Entwicklung von hochtechnisierten Produkten ausgesprochen reichen stärken. Mensch bei autonomen Systemen wesentlich weniger Fehler als • Der Schutz von Menschen vor Schaden (psychisch/körperlich/ außergewöhnlich. 2. [Tec_C_Emissionen] Das autonome Fahren wird den Verkehrs- beim Menschen selbst. Daneben verursacht die Angst vor Ha- monetär) durch autonome Systeme muss sichergestellt werden. fluss optimieren und Staus reduzieren, sodass der Kraftstoffver- ckerangriffen großen gesellschaftlichen Widerstand gegen auto- Aufgrund der Mannigfaltigkeit der Situationen, die ein autonomes Mit dem autonomen Fahren sind hohe Ziele verknüpft: die Politik brauch und die Emissionen verringert werden.[4] nomes Fahren. Manche Menschen möchten auch auf das eigene System gegebenenfalls bewältigen muss, liegt hier eine wesent- setzt vor allem auf das autonome Fahren als umfassendes Sicher- Fahrerlebnis nicht verzichten. liche Herausforderung für die Forschung. [Tec_R_Verkehrsteilneh- heitssystem, ohne dass die Ziele aus dem Weißbuch zum Transport Risiken 2. [Ges_R_Datenschutz] Der Datenschutz spielt beim autonomen merschutz, Ges_R_Automatisierung] der Europäischen Kommission[1] nicht umsetzbar wären: "By 2050, 1. [Tec_R_Sicherheit] Die mangelnde Sicherheit des Systems, sowohl Fahren eine große Rolle. Fahrbare Computer sammeln beispiels- • Die Vernetzung zwischen unterschiedlichen Anwendungsgebieten move close to zero fatalities in road transport." Da 90 % aller töd- in Bezug auf Verlässlichkeit, als auch den Schutz vor unbefugtem weise Daten über die Passagiere, um das Infotainmentsystem muss gefördert werden, um in der Praxis erfolgreiche Methoden lichen Unfälle auf der Straße die Folge von Schläfrigkeit, Ablenkung, Zugriff Dritter, ist ein zentrales Risiko für das autonome Fahren. auf deren Bedürfnisse anzupassen. Auch Kameraaufnahmen der und Techniken auszutauschen. [Tec_C_Errungenschaft] Trunkenheit, Schnelligkeit und anderen menschlichen Faktoren Die verteilten Systeme im Auto sowie deren Kommunikation un- Umgebung werden gesammelt. Es kommen Fragen auf, wem die • Forschung zu aktiv Kontrolle übernehmenden Systemen ist nicht sind, liegt es nahe, auf autonome Systeme zu setzen, um die Un- tereinander und mit anderen Fahrzeugen müssen abgesichert Daten gehören, wer Zugriff darauf hat und welche Daten wo ge- ausschließlich technologisch getrieben, sondern in die Gesell- fallgefahr – insbesondere schwerwiegender Unfälle mit Personen- werden, sodass weder Zugriff auf sensible Daten noch auf Kom- speichert werden.[8] schaft eingebettet. Die Forschung muss Methoden entwickeln, schaden – systematisch zu verringern. Die bisherigen Erfahrungen ponenten möglich ist.[5] 3. [Ges_R_Ethik] Zudem müssen ethische Fragen behandelt werden. um in einen effektiven Dialog mit der Gesellschaft treten zu kön- mit (teil-)autonomen Fahrzeugsystemen lassen allerdings nicht da- 2. [Tec_R_Verkehrsteilnehmerschutz] Kommen vermehrt Verkehrs- Menschenleben dürfen in Europa nicht gegeneinander aufge- nen, der dazu dient, die gesellschaftlichen Rahmenbedingungen rauf schließen, dass eine Reduzierung fataler Unfälle auf Null rea- teilnehmer (psychisch/ körperlich/monetär/...) zu Schaden, ins- wogen werden. Es stellt sich die Frage, welche Regeln im Falle zu erheben und zu beachten. [Ges_R_Ethik, Ges_R_Automatisierung] listisch erreichbar sein wird. besondere im Vergleich zum Ist-Stand, so ist dies nicht verant- unausweichlicher Kollisionen den größten Rückhalt in der Bevöl- • Es müssen Konzepte entwickelt werden, die eine demokratisch wortbar; genauer ist wohl nur eine erhebliche Verbesserung des kerung finden. legitimierte Kontrolle der autonomen Systeme ermöglicht. Dazu Aus der Sicht der Wissenschaft sind mit dem autonomen Fahren Verkehrs- teilnehmerschutzes gesellschaftlich vertretbar, um das 4. [Ges_R_Automatisierung] Nicht zuletzt fürchten viele Menschen ei- gehört vor allem eine Verfügbarmachung und Analyse der anfal- nicht nur ingenieurwissenschaftliche und informationstechnische empfundene Zusatzrisiko auszugleichen, das mit dem Gefühl bei nen Arbeitsplatzverlust durch die zunehmende Automatisierung. lenden Daten.[10] Eine rein marktwirtschaftliche Auswertung und Fragestellungen zu bearbeiten. Diese sind zwar aus Sicht der Mach- der Abgabe der Kontrollhoheit einhergeht. Diese Ängste müssen adressiert und neue Perspektiven geboten Nutzung der Daten durch wenige industrielle Stakeholder ist ge- barkeit des autonomen Fahrens von essentieller Bedeutung, aber werden. Ebenso kann eine allgemeine Aversion dagegen beste- sellschaftlich nicht wünschenswert. [Ges_C_Staatliche_Steuerung, auch psychologische, ethische, philosophische und juristische Fra- Gesellschaftliche Chancen und Risiken hen, gewisse Abläufe automatisieren zu lassen. Ges_R_Rebound_Effekt] gen sind nicht von geringerer Bedeutung. Chancen 5. [Ges_R_Teilautomatisierung] Neue Gefahren ergeben sich zudem 1. [Ges_C_Verkehrssicherheit] Schon heute machen Assistenzsyste- aus den Zuständen der Teilautomatisierung von Fahrzeugen. Wirtschaft und Industrie Sowohl die Fahrzeughersteller als auch die Fahrzeugführer verbin- me das Fahren sicherer und retten Leben. Da mehr als 90% der Sind den Menschen die unterschiedlichen Stufen der Autono- • Die Verlässlichkeit der Systeme muss gegebenenfalls vor die den mit dem autonomen Fahren neben der Sicherheit moderne Verkehrsunfälle auf menschliches Fehlverhalten zurückzuführen misierung und ihre Auswirkungen nicht bekannt, so kann es zu Funktionalität gestellt werden. Bei der Entwicklung von Software und besonders komfortable Fahrtechnik sowie das Bedienen neuer sind, kann die Unfallrate durch automatisiertes Fahren reduziert gesteigertem Unfallrisiko kommen. Dies ist beispielsweise der für autonome Systeme müssen die Qualitätsstandards gesteigert

[1] https://ec.europa.eu/transport/sites/transport/files/themes/strategies/doc/2011_ [4] Strategie automatisiertes und vernetztes Fahren. Berlin, Bundesministerium für Ver- [6] Position: Automatisiertes Fahren - rechtliche Hürden beseitigen. München, vbw Vereini- [8] Position: Automatisiertes Fahren - Datenschutz und Datensicherheit. München: vbw white_paper/white-paper-illustrated-brochure_en.pdf, S. 9 kehr und digitale Infrastruktur, 2015, S. 10 gung der Bayerischen Wirtschaft e.V. 2018, S. 5f. Vereinigung der Bayerischen Wirtschaft e.V. 2018, S. 3 - 24 [2] Die jeweiligen Chancen und Risiken sind mit einer eindeutigen ID versehen, wobei [5] Position: Automatisiertes Fahren - Datenschutz und Datensicherheit. München: vbw [7] Strategie automatisiertes und vernetztes Fahren. Berlin, Bundesministerium für Ver- [9] https://www.umweltpakt.bayern.de/nachhaltigkeit/fachwissen/347/der-rebound-effekt Tec_C für “Technische Chance”, Tec_R für “Technisches Risiko”, Ges_C für “Gesell- Vereinigung der Bayerischen Wirtschaft e.V. 2018, S. 29 - 33 kehr und digitale Infrastruktur, 2015, S. 9 schaftliche Chance” und Ges_R für “Gesellschaftliches Risiko” stehen. [3] Position: Datenschutz, IT-Sicherheit und Haftung bei automatisierten Systemen. München, vbw Vereinigung der Bayerischen Wirtschaft e.V. 2017, S. 13f.

52 | ARBEITSKREISE ARBEITSKREISE | 53 ARBEITSKREISE

werden, da Menschenleben auf dem Spiel stehen. Beim Einsatz von Methoden der Künstlichen Intelligenz besteht großer For- schungsbedarf in der Absicherung der zugrundeliegenden datengetriebenen Modelle. [Tec_R_Sicherheit, Ges_R_Automatisierung, Ges_R_Fehlertoleranz_Akzeptanz] • Der Schutz der Menschen muss im Produkt soweit möglich sichergestellt werden.[11] [Tec_R_Verkehrsteilnehmerschutz, Ges_R_Auto- matisierung] • Die Wirtschaft muss Methoden und Werkzeuge nutzen, um in einen Dialog mit der Gesellschaft zu treten, um die gesellschaftlichen Rahmenbedingungen bei der Entwicklung aktiv Kontrol- le übernehmender Systeme zu erheben und zu beachten.[12] [Ges_R_Ethik, Ges_R_Automatisierung]

Ethik, Jurisprudenz und Gesellschaft • Es ist ein juristisch bindender Vorschriftenkatalog von Sicherheits- und Schutzmaßnahmen zu entwickeln, der bei der Entwicklung autonomer Systeme zu beachten ist [Tec_R_Sicherheit, Tec_R_Ver- kehrsteilnehmerschutz, Ges_R_Fehlertoleranz_Akzeptanz] • Staatliche Institutionen sollen durch richtungsweisende Gesetzge- bung ihrer Kontrollaufgabe im Bezug auf autonome Systeme gerecht werden. [Ges_C_Staatliche_Steuerung, Ges_R_Rebound_Effekt] • Die Gesellschaft muss über Chancen und Risiken autonomer Sys- teme aufgeklärt werden. [Ges_R_Fehlertoleranz_Akzeptanz] • In Bezug auf den Datenschutz müssen geeignete Gesetze verab- schiedet und Regelungen aufgestellt werden, um das Vertrauen der Gesellschaft nicht zu verlieren. Es gilt zu klären, wann, wo und welche Daten gesammelt und an wen gesendet werden. Transpa- renz ist in diesem Zusammenhang essentiell. [Ges_R_Datenschutz]

Fazit Autonome Systeme eröffnen vielfältige Chancen, bergen aber auch Risiken für Technologie und Gesellschaft. Daraus resultierende Handlungsempfehlungen für Lehre, Forschung und Entwicklung, Wirtschaft und Industrie und auch Ethik, Jurisprudenz und Gesell- schaft haben wir in diesem Beitrag dargelegt.

AUTOREN

Ramin Tavakoli Kolagari (TH Nürnberg – ZD.B) Katja Auernhammer (TH Nürnberg – ZD.B) Anja Bodenschatz (TU München – ZD.B) Konrad Doll (HAW Aschaffenburg – ZD.B) Viktor Kress (HAW Aschaffenburg – ZD.B) Jürgen Mottok (OTH Regensburg – ZD.B) Matthias Uhl (TU München – ZD.B)

[10] (Rechts-)Sicher Fahren mit Autopilot. Berlin, Verbraucherzentrale Bundesverband e.V. [12] https://www.welt.de/wirtschaft/article146407129/Wen-soll-das-autonome-Auto-lie- 2016, S. 3. ber-ueberfahren.html [11] Problem der kulturellen Phasenverschiebung “cultural lag”, wonach die technologische Dynamik den politisch gesellschaftlichen-Regelungsprozessen mehrere Schritte voraus ist, vgl. W.F. Ogburn, Social Change With Respect to Culture and Original Nature, New York 1922.

54 | ARBEITSKREISE 55 Daten zu unterstützen. Neben einer verbesserten Diagnostik be- Handlungsempfehlungen ARBEITSKREISE steht das Potenzial zu einer vorzeitigen Erkennung von Krankheiten Neue Technologien, wie Medical Wearables, und Verfahren, wie und zur genauen Vorhersage von zukünftigen Krankheitsverläufen KI-basierte Bestimmung von Gesundheitsmarkern und Entschei- nicht nur in Kliniken mit hoher medizinischer Expertise. Mit Hilfe von dungsunterstützung, können die Gesundheitsvorsorge und medizi- KI-Algorithmen ist es möglich, quantitative Messungen aus Bildern zu nische Betreuung von Patienten revolutionieren. Bayern hat durch extrahieren, sogenannte Imaging-Biomarker. Diese können, ähnlich die hohe Anzahl von KMUs mit Bezug zu medizinischer Technolo- zu einem Laborbericht, dem Arzt zusätzlich zu den entsprechenden gie sowie die medizinischen Forschungseinrichtungen bereits einen Normbereichen zur Verfügung stehen. Während in der Forschung sichtbaren Vorsprung. Um die führende Position Bayerns langfristig DIGITAL HEALTH bereits Verfahren entwickelt wurden, die quantitative Messungen zu festigen sind eine Reihe von Maßnahmen notwendig. mit hoher Genauigkeit ermöglichen, ist die klinische Umsetzung POSITIONSPAPIER noch ungelöst. Neben der Extraktion von quantitativen Messungen Der Transfer von neuen Methoden und Technologien in alltägliche werden auch Algorithmen eingesetzt um Diagnosen direkt vorherzu- medizinische Versorgung verläuft schleppend. Oft wird die Umset- sagen. Solche Systeme würden nicht in Isolation, sondern immer in zung durch die Arbeitsbelastung des medizinischen Personals ge- Umfassende und individualisierte Gesundheitsversorgung auf dem Deutschland muss den Rückstand in Bereichen der medizinischen starker Zusammenarbeit mit medizinischen Experten zur Entschei- bremst. Prozessumstellungen, Erprobung von Innovationen, sowie höchsten Niveau ist eine zentrale Herausforderung des 21. Jahrhun- Informatik und Medizintechnologien gegenüber anderen Ländern dungsunterstützung eingesetzt werden. Eine große Herausforde- Wirtschaftlichkeitsanalysen müssen verstärkt gefördert werden um derts. Lebensqualität und die Exzellenz medizinischer Versorgung aufholen. Die Medizininformatik-Initiative des BMBF ist dabei ein rung hierbei ist es, die getroffene Entscheidung nachvollziehbar zu die Lösungsentwicklungen voran zu bringen. Ein schnelles, einfaches werden schon heute als essentielle Standortfaktoren im weltweiten wichtiger Schritt, muss aber weiter ausgebaut werden. Bayern machen. Die Entwicklung eines besseren Verständnisses KI-basierter und fokussiertes Förderverfahren „Translationaler Forschung und Vergleich von Regionen und Städten benutzt. Kaum ein anderer Be- hat durch verschiedenen Maßnahmen eine führende Position im Systeme sowie transparenter Systeme für die Entscheidungsunter- Erprobung“ muss geschaffen werden, um die transdisziplinäre Zu- reich hat einen so weitreichenden Einfluss auf die Gesellschaft und Vergleich zu anderen Bundesländern. Die Modellregion digitale stützung ist deshalb notwendig. sammenarbeit in einem Rahmen von ca. 100k€ pro Vorhaben zu Wirtschaftsleistung wie die Gesundheitsversorgung: Die steigende Gesundheitswirtschaft Franken, Medical Valley, sowie die Unter- unterstützen. Die bayerische Wissenschaftsförderung, besonders Lebenserwartung trägt zum rapiden Anstieg chronischer Krank- stützung der Ausstattung von dezentralen Krankenhäusern und Arbeitsprozesse: Die Digitalisierung erlaubt es Arbeitsprozesse zu durch die Themenplattformen des ZD.B, ist anerkennenswert. Jedoch heiten bei, wie Hypertonie, Fettleibigkeit, Diabetes, und Herz-Kreis- niedergelassenen Ärzten sind wesentliche Elemente einer nach- optimieren, insofern diese sinnvoll in die bestehenden Routinen in- ist im Bereich der Medizin die Förderung von jungen Unternehmen lauferkrankungen, die neben der eingeschränkten Lebensqualität haltigen bayerischen Gesundheitspolitik. Dennoch müssen auch tegriert werden. Digitale Prozesse und medizinische Apps können und universitärer Forschung mit mehr als 50% der Gesamtprojekt- auch wesentlich zu den Gesundheitskosten beitragen. Daneben in Bayern weitere Anstrengungen unternommen werden, um den Diagnose, Therapie, Überwachung von Krankheiten, und Rehabi- kosten und Laufzeiten von drei Jahren notwendig, um wesentliche entstehen zusätzliche Herausforderungen durch anhaltenden All- Standortvorteil für die zahlreichen medizin-nahen KMU sowie für litation unterstützen. Die automatisierte Analyse erfasster Daten, Innovationen, Promotionsarbeiten oder den Transfer zu erreichen. tagsstress, Bewegungsmangel und Umwelteinflüsse, die zu Schlaf- die Wissenschaft zu stärken. Insbesondere ist für Bayern die wei- z.B. Krankheitshistorie, Vitalparameter, Bewegungsdaten, erlaubt störungen, Depression oder Burn-out führen und so die volkswirt- tere Investition in die Bereiche Prävention, klinische Diagnostik, es relevante Informationen zu filtern, visualisieren und ermöglicht Die Einbindung von Bürgern in die Entwicklung der Medizin, Ge- schaftliche Leistungsfähigkeit einschränken. Beispielsweise gilt sowie Telemedizin und außerklinische Patientenunterstützung somit die direkte Nutzung für Kliniker und Patienten. Die rasche sundheitsvorsorge und Technologien im Rahmen von Citizen Sci- Bewegungsmangel inzwischen als eigenständiger Risikofaktor für sinnvoll, um langfristig einen substantiellen Vorsprung zu halten. Übertragung medizinischer Daten und objektiver Parameter kann ence-Initiativen kann neben der Transparenz und Sichtbarkeit des die Entstehung und Chronifizierung nicht ansteckender Krankheiten das medizinische Team in wichtigen Entscheidungsprozessen sowie wissenschaftlichen Fortschritts und der klinischen Arbeitsprozesse und ist der viertgrößte Risikofaktor für Todesfälle weltweit. Zur Ma- Prävention: Die Krankheitsprävention verändert sich zusehends: der Patientenberatung unterstützen. Die Qualitäts- und Leistungser- auch die Gesundheitskompetenz jedes Einzelnen steigern. Ein For- ximierung der Lebensqualität von Bürgern und Patienten, flächen- Waren vor einigen Jahren die ärztlichen Untersuchungen noch die fassung sowie Dokumentation sind wichtige Elemente, die mittels di- mat für lokale Projekte ist wünschenswert, bei denen sich Bürger deckender medizinischen Versorgung und kontrollierbarer Kosten, alleinige Maßnahme vieler, so sind heute eine Vielzahl von mobi- gitaler Prozesse einen schnellen, disziplinübergreifenden und stand- engagieren können, von Abendvorträgen, offenen Laboren, bis zur können neue Diagnostik- und Interventionstechnologien einen we- len Technologien, einschließlich Apps, Fitnesstracker, und weiteren ortunabhängigen Informationsaustausch ermöglichen können. geführten Analyse eigener Daten mittels partizipativer Entwicklungs- sentlichen Beitrag leisten. Wearables im Konsumentenmarkt verfügbar. Durch personenzent- Intelligente Vernetzung zwischen verschiedenen Einrichtungen und ansätze. Bereits im schulischen Umfeld muss das Bewusstsein für rierte Daten können sowohl die Gesundheitsförderung und Präven- Patienten (elektronische Patientenakte) bietet die Chance bestehen- Gesundheit und die Möglichkeiten neuer medizinischen Verfahren Die ‘All of us’-Initiative in den USA, sowie die Investitionen in Bio- tion neuer Erkrankungen unterstützt werden, als auch ein wichtiger de Abläufe zu beschleunigen und die Transparenz für alle Beteiligten und Technologien durch kurrikulare Ergänzungen geschaffen wer- technologie in China sind Beispiele eines globalen Interesses aller Beitrag zur Sekundärprävention geleistet werden, indem z.B. früh- zu erhöhen. Jedoch gibt es in Deutschland vielfältige Barrieren, z.B. den. großer Wirtschaftsräume, wissenschaftliche Erkenntnisse mög- zeitig in die Krankheitsprozesse chronisch Erkrankter eingegriffen die fehlende Einbindung des Personals in Prozessumstellungen, die lichst schnell in optimierte Behandlungskonzepte zu überführen. Im werden kann. Während die mobilen Technologien auf diese Weise systematisch und gezielt unter Einbindung aller Beteiligten abgebaut Die Kl-basierten Analysemethoden von medizinisch relevanten In- Vordergrund der Initiativen steht oft die Personalisierung der me- die Bewegung und Gesunderhaltung für einen Teil der Nutzer posi- werden müssen. formationen erfordert große Datenbestände, wie sie aktuell kaum dizinischen Versorgung und Prävention basierend auf großen, ver- tiv beeinflussen, sind die Funktionalitäten häufig nicht validiert und öffentlich verfügbar sind. Der Datenmangel motiviert das aktuelle knüpften medizinischen Datenbanken. US-Konzerne, insbesondere schaffen in den meisten Fällen keinen nachhaltig gesunden Lebens- Nachsorge, Telemedizin und Außerklinische Unterstützung: Sammelinteresse von Gesundheits-Apps. Es besteht die Notwen- Apple und Google, haben das Potential von medizinischen Daten stil in den relevanten Zielgruppen, z.B. für Bewegungsmangel oder Körpergetragene Sensor-Systeme, oder Medical Wearables, finden digkeit, ein Register für wiederholt nutzbare medizinische Datenbe- bereits seit einigen Jahren erkannt und nutzen ihre Marktposition, Übergewicht. Kritische Barrieren müssen abgebaut werden: Inno- zunehmend Einsatz in der Rehabilitation und Nachsorge. Momentan stände zu schaffen, welches auf medizinisch-ethischen Grundlagen um Nutzerdaten mit Hilfe von mobilen Geräten und Wearables zu vative und beratende Funktionen müssen geschaffen werden, um konzentrieren sich Medical Wearables auf Anwendungen im klini- aufbauend den Zugriff auf Daten schafft. Eine Vereinfachung der sammeln, einschließlich der persönlichen Patientenakte sowie Ver- Nutzern in ihrem individuellen Alltag situativ geeignete Unterstüt- schen Umfeld, z.B. Spital oder Rehabilitations-Zentrum. Die aktuelle Sekundärnutzung von klinischen Daten für Forschungszwecke ist haltens- und Umweltdaten. Für die Wissensgewinnung und Patien- zung zu bieten. Bereits heute lassen sich einzelne komplexe Alltags- Forschung greift Ansätze der Telemedizin auf, um Patienten auch erforderlich. Ebenso ist eine Teilregulierung der Gesundheits-Apps tenbetreuung ermöglichen die Datenintegration und neue Metho- situationen durch Integration von virtuellen Daten, z.B. Kalender, außerklinisch zu unterstützen. Dabei werden zunehmend energieef- nötig, um Bürgern mehr Sicherheit und Transparenz hinsichtlich den der künstlichen Intelligenz (KI) umfassende Interpretationen mit Orts- und Bewegungsinformationen für situative Verhaltens- fiziente Sensor-Systeme in Accessoires wie Brillen, Armbänder oder der Ausnutzung ihrer Daten zu verschaffen. Im Rahmen der baye- zu individuellen Routinen, Risikofaktoren und Krankheitsverläufen. empfehlungen auswerten, beispielsweise bei der Speisenauswahl Schuhen integriert. Somit dringen die Medical Wearables in den All- rischen Initiative zur elektronischen Patientenakte ist die Untersu- Jedoch lassen sich die Informationen aus den Daten auch für kom- im Restaurant oder im Supermarkt. Für eine durchgehende, situ- tag von Patienten ein ohne diese zu stören.[2] Mit Medical Wearables chung innovativer Lösungen für die Regulierung und Datennutzung merzielle Zwecke nutzen. Besonders bei privatwirtschaftlichen Lö- ative Alltagsunterstützung sind sowohl geeignete Technologien für können Gesundheitsmarker, wie beispielsweise die Herzratenvaria- von großem Bedarf. Außerdem muss ein durchgehendes digitales sungen erscheint das Missbrauchsrisiko hoch, beispielsweise um die Verhaltensbeobachtung, als auch eine Weiterentwicklung der bilität, kontinuierlich aus Sensordaten gewonnen werden. Ebenfalls Patientenmanagement geschaffen werden, um lückenlose An- durch individuelle Risikoanalysen die Versicherungsprämien oder KI-Methoden für maschinelles Kontextbewusstsein notwendig. kann das Bewegungsverhalten analysiert werden. Somit schafft schlussversorgung und Nachsorge zu ermöglichen. Vergünstigungen zu steuern. Das Potential für Datenmissbrauch die Kombination aus Wearables und KI-basierter Datenanalyse die steigt auch durch Konzeptionsfehler: Beispielsweise ist bei vielen Klinische Diagnostik: Die Aufgabe des medizinischen Personals Grundlage für personalisierte Trainings- und Therapie-Empfehlun- Smartphone-Apps unklar, wo die Patientendaten gespeichert wer- ist es, vielfältige Informationen aus multimodalen Datenquellen zu gen. Neben Medical Wearables bieten gebäudeintegrierte Systeme den und wer dadurch physischen Zugriff erlangen kann, oder es einem konsistenten Gesundheitsüberblick eines Patienten zusam- und Technologien die Chance Patienten im häuslichen Umfeld zu AUTOREN werden Eingabefelder mit Suchmaschinen verknüpft.[1] Bei dem menzufügen. Dies ist eine große Herausforderung, da es sich hierbei unterstützen. Zudem wird untersucht, wie sich Gesundheitsmarker Oliver Amft (FAU, Lehrstuhl für Digital Health – ZD.B) schnellen Fortschritt der Technologie muss es das höchste Ziel blei- teilweise um hochkomplexe Daten handelt. Allein die medizinische zu einem effizienten digitalen Patientenmanagement kombinieren Adrian Derungs (FAU, Lehrstuhl für Digital Health – ZD.B) ben, die Privatsphäre des Einzelnen zu schützen sowie das Recht Bildgebung produziert Bilder in einer höheren Geschwindigkeit als lassen. Die Privatsphäre und die Datensicherheit müssen dabei für Luis I. Lopera Gonzalez (FAU, Lehrstuhl für Digital Health – ZD.B) auf Selbstbestimmung zur Datenfreigabe mit adäquaten Mitteln zu radiologische Expertise zur manuellen Interpretation zur Verfügung den Patienten stets verständlich und kontrollierbar bleiben. Christian Wachinger (LMU München, Klinikum – ZD.B) ermöglichen. steht. Die computergestützte Diagnose ist folglich dringend notwen- Nina Rohrbach (TUM, Lehrstuhl für Bewegungswissenschaft – dig um Ärzte bei der Verarbeitung von komplexen medizinischen ZD.B) [1] Beispiel Gesundheits-App Vivy: https://www.heise.de/newsticker/meldung/Daten- [2] Überblick in: Amft, O. (2018). How Wearable Computing Is Shaping Digital Health. IEEE schutzdebatte-um-neue-Gesundheits-App-Vivy-4169288.html Pervasive Computing, 17(1), 92–98.

56 | ARBEITSKREISE ARBEITSKREISE | 57 SYMPOSIUM

AKTEURE SEITE 7 ZD.B SYMPOSIUM 2018: AUFTAKT DES 1. NETZWERKKONGRESSES DES ZD.B RESEARCH INCUBATORS

FRAUENCHIEMSEE, 25. & 26. OKTOBER 2018

Nach dem erfolgreichen Aufbau der ZD.B Initiativen für die Wissenschaft haben sich in diesem Jahr erstmalig alle wissenschaftlichen Akteure des Zentrum Digitalisierung.Bayern zu einem Symposium getroffen, um zwei Tage des wissenschaftlichen Austauschs und voller inspirierender Diskussionen im breiten Themenfeld der Digitalisierung im Kloster Frauenwörth auf der Fraueninsel im Chiemsee zu verbringen.

ARBEITSKREISE Zu den über 70 Gästen zählten die neu berufenen ZD.B ProfessorInnen, ZD.B NachwuchsforschungsgruppenleiterInnen, ZD.B Fellows sowie Promovierende der verschiedenen ZD.B Forschungsgruppen und des BayWISS Verbundkollegs Digita- SEITE 40 lisierung. Daneben fand auch die Preisverleihung des ZD.B Promotionspreises an die diesjährigen drei Preisträgerinnen und Preisträger statt.

Das Programm des ersten Symposiums zeichnete sich vor allem durch vielseitige Beiträge aus der Wissenschaft, anregende Diskussionen unter allen Teilnehmenden und – das Herzstück der Veranstaltung – die interaktiven Posterpräsentationen unterschiedlichster digitalisierungsrelevanter Forschungsprojekte der Promovierenden aus.

Die Themenbereiche deckten ein breites Feld von IT-Sicherheit & Datenschutz über Mobility, Energieeffizienz, Virtual Reality und Mensch-Maschine-Interaktion bis hin zu interpretierbarem Machine Learning ab. Für weitere Impulse sorgte die Keynote von Professorin Ursula Münch zum Thema „Auswirkungen der digitalen Transformation auf Politik und Gesellschaft“. Im abschließenden Closing Panel „From PhD to Startup – Insights from Entrepreneurs“ zeigten drei UnternehmerInnen alternative Karrierewege nach der Promotion auf und teilten ihre Erfahrungen zum Thema Entrepreneurship mit dem Publikum.

Das ZD.B Symposium soll nun als jährlicher Netzwerkkongress des ZD.B Forschungsinkubators verstetigt werden. Unser SYMPOSIUM Ziel ist es, die unterschiedlichen Akteure der ZD.B Initiativen für die Wissenschaft im Rahmen unserer interdisziplinären Forschungs- und Kollaborationsplattform regelmäßig zusammenzubringen, um die Themenvielfalt der Digitalisierung ge- SEITE 58 meinsam adressieren zu können. Dies und noch viel mehr wurde nun bei herbstlicher Insel-Idylle in die Tat umgesetzt, sodass alle Teilnehmenden die Rückreise aufs Festland mit neuen Ideen und Kontakten im Gepäck antreten konnten.

Im Folgenden finden Sie Impressionen der Veranstaltung, das Programm des Symposiums sowie die eingereichten Abstracts der Promovierenden für die Posterpräsentationen.

58 | SYMPOSIUM SYMPOSIUM | 59 SYMPOSIUM IMPRESSIONEN

Zu jeder Tageszeit beeindruckte das Kloster Frauenwörth die Teilnehmenden des Symposiums mit einem einzigartigen Ausblick auf Chiemsee und Alpenland. Über 70 Teilnehmer, darunter Professoren, Nachwuchsgruppenleiter und Promovierende aus ganz Bayern, tauschten sich zu ihren Forschungsprojekten und Ideen aus.

Gründungspräsident und wissenschaftlicher Geschäfts- führer Professor Dr. Dr. h.c. Manfred Broy betont in seiner Eröffnungsrede den interdisziplinären Charakter des ZD.B.

Der diesjährige ZD.B Promotionspreis wurde im Rahmen der Veranstaltung feierlich überreicht. SYMPOSIUM IMPRESSIONEN

Durch knackige wie informative "Blitz-Intros" lockten die Promovierenden interessierte Zuhörer zu ihrer Posterwand, um ihr Promotionsprojekt dort im Detail vorzustellen. Komplexe Forschungsprojekte aus Bereichen wie Mensch-Maschine Interaktion, Datensicherheit und Mobility wurden gekonnt visualisiert und damit auch für Fachfremde zugänglich gemacht.

Oben: Keynote-Speakerin Prof. Dr. Ursula Münch Unten: Das Experten-Panel diskutiert über "Digitisation and the Society of Tomorrow" SYMPOSIUM PROGRAMM

16:00 UHR Ankunft im Kloster Frauenchiemsee 09:00 UHR Begrüßung TAG 17:30 UHR Get-Together TAG 09:15 UHR 03: Let's talk about: Intelligent and Automated Systems Prof. Dr. Konrad Doll – HAW Aschaffenburg 18:00 UHR Abendessen Prof. Dr. Andreas Festag – TH Ingolstadt Prof. Dr. Sascha Hauke – HAW Landshut Prof. Dr. Martin Matzner – Universität Erlangen-Nürnberg 00MI. 24. OKT. 2018 02FR. 26. OKT. 2018 Prof. Dr. Dominik Herrmann – Universität Bamberg 10:15 UHR Kaffeepause 10:00 UHR Begrüßung ZD.B und Grußworte StMWK Prof. Dr. Dr. h.c. Manfred Broy – Gründungspräsident und wissenschaftlicher 10:30 UHR FELLOWS COACHING OPEN SPACE EVENT TAG Geschäftsführer ZD.B Session C Dr. Peter Müller – Amtschef des Bayerischen Staatsministeriums für Wissenschaft und Kunst (StMWK) 10:30 UHR Verleihung des ZD.B Promotionspreises 11:15 UHR FELLOWS COACHING OPEN SPACE EVENT Dr. Hanna Schneider – LMU: Hauptpreis Session D 01DO. 25. OKT. 2018 Dr. Jonas Biehler – TUM: Preis für die beste interdisziplinäre Arbeit Dr. Benjamin Schleich – FAU: Preis für die beste anwendungsorientierte Arbeit Mittagspause 10:45 UHR Opening Keynote 12:00 UHR Prof. Dr. Ursula Münch 13:00 UHR FELLOWS COACHING OPEN SPACE EVENT Universität der Bundeswehr München Akademie für Politische Bildung Session E 11:30 UHR Blitz-Intro Postersession A 13:45 UHR 04: Let's talk about: From Computational Modelling to Design for the Wellbeing 12:00 UHR Mittagspause Dr. Christian Wachinger – LMU München Prof. Dr. Björn Schuller – Universität Augsburg Postersession A 13:00 UHR Prof. Dr. Oliver Amft – Universität Erlangen-Nürnberg Prof. Andreas Muxel – HAW Augsburg A1 A2 A3 A4 A5 14:45 UHR Kaffeepause

Closing Panel 01: Let's talk about: Digitisation and the Society of Tomorrow 15:00 UHR 14:00 UHR From PhD to Startup – Insights from Entrepreneurs Dr. Daniel Schnurr – Universität Passau Dr. Matthias Uhl, vertreten durch Dr. Gari Walkowitz – TU München Prof. Dr. Nicholas Müller – HAW Würzburg-Schweinfurth 15:45 UHR Wrap-Up und Verleihung der Posterpreise Prof. Dr. Florian Alt – HAW München 16:00 UHR Ende des Symposiums 15:00 UHR Kaffeepause 15:30 UHR 02: Let's talk about: Human-centred Engineering Prof. Dr. Jörg Müller – Universität Bayreuth Prof. Dr. Albrecht Schmidt – LMU München Prof. Dr. Jens Grubert – HAW Coburg Dr. Raphael Wimmer – Universität Regensburg Dr. Janet Siegmund – Universität Passau 16:30 UHR Blitz-Intro Postersession B 17:00 UHR Postersession B

B1 B2 B3 B4 B5 Moderation: Eva Wolfangel

Veranstaltungsort: Abendessen 18:00 UHR Großer Seminarraum 19:00 UHR Open-Space-Pitches & Networking Kleine Seminarräume 64 | SYMPOSIUM Essens- und Gemeinschaftsräume SYMPOSIUM | 65 SYMPOSIUM PROGRAMM

GRUPPENEINTEILUNG FÜR TALKSHOWS GRUPPENEINTEILUNG FÜR POSTERSESSIONS

01: LET’S TALK ABOUT: SESSION A Digitisation and the Society of Tomorrow Gruppe A1 (Raum „Alan Turing“) A1 Alice Baird Reinforced Soundscape Generation for Wellbeing via Multimodal Individual Dr. Daniel Schnurr – Universität Passau: Data Neutrality & Open Access: Coherent Economic Policies for the Monitoring Digital Economy (Nachwuchsforschungsgruppe) Pascal Knierim Mixed Realities in Everyday Life Dr. Matthias Uhl, vertreten durch Gari Walkowitz – TU München: Ethics of Digitization Viktorija Paneva Dexterous Interaction with Levitating Interfaces (Nachwuchsforschungsgruppe) Florian Bockes User Centered workflow analysis for innovative Systems in public administration Prof. Dr. Nicholas Müller – HAW Würzburg-Schweinfurth: Sozioinformatik und gesellschaftliche Aspekte der Digitalisierung Gruppe A2 (Raum „Margaret Hamilton“) Prof. Dr. Florian Alt – HAW München: Biometrics – Leveraging Behavioral Biometrics Beyond Security to Both A2 Markus Zoppelt IT-Security for Autonomous and Connected Driving Secure and Personalize Interactive Ubiquitous Computing Devices (Nachwuchsforschungsgruppe) Markus Ring Data Mining for Network-based Intrusion Detection Katja Auernhammer Security for Machine Learning in the Context of Autonomous Vehicles 02: LET’S TALK ABOUT: Viktor Kress Vehicle-Based Intention Detection of Vulnerable Road Users Human-centred Engineering Gruppe A3 (Raum „Konrad Zuse“) Prof. Dr. Jörg Müller – Universität Bayreuth: Serious Games / Applied Games A3 Clemens Birkenmaier A realistic and patient specific computational model of blood viscosity Prof. Dr. Albrecht Schmidt – LMU München: Human-Centered Ubiquitous Media and flow induced coagulation Prof. Dr. Jens Grubert – HAW Coburg: Mensch-Maschine-Interaktion im Internet der Dinge Emanuel Marx Optimization of production processes using Process Mining Dr. Raphael Wimmer – Universität Regensburg: Physical Affordances for Digital Media and Workflows Kevin Pfeffel Measuring cognitive load levels during active sport sessions and the use for (Nachwuchsforschungsgruppe) digitalization Dr. Janet Siegmund – Universität Passau: PICCARD: Improving Collaboration and Communication of Matthias Stierle Bridging the gap between process analysis and process automation – Developers (Nachwuchsforschungsgruppe) An evidence-based process automation approach

Gruppe A4 (Raum „Grace Hopper“) A4 03: LET’S TALK ABOUT: Jonas Schlund Blockchain-Based Orchestration in Decentralized Power Systems Victoria Fast The Collection of Personal Data: Competitive and Behavioral Effects Intelligent and Automated Systems Gari Walkowitz Experimental Studies on Filling AI-Induced Retribution Gaps through Scapegoat Punishment Prof. Dr. Konrad Doll – HAW Aschaffenburg: Kooperative automatisierte Verkehrssysteme Anja Bodenschatz Behavioral Implications of Human-Machine-Interactions – Experimental Prof. Dr. Andreas Festag – TH Ingolstadt: Fahrzeugsicherheit und Car2X-Kommunikation Evidence on the Resolution of Ethical Dilemmas by Decision Randomisation Prof. Dr. Sascha Hauke – HAW Landshut: Intelligente Energienetze Prof. Dr. Martin Matzner – Universität Erlangen-Nürnberg: Digital Industrial Service Systems Gruppe A5 (Raum „Edsger Dijkstra“) Prof. Dr. Dominik Herrmann – Universität Bamberg: Privatsphäre und Sicherheit in Informationssystemen A5 Christoph Molnar Interpretable Machine Learning Philipp Dufter Effective usage of multilingual data in Natural Language Processing Sarah Wunderlich Self-adapting Online Analysis of Host-based Security-critical Events in 04: LET’S TALK ABOUT: SME Company Networks From Computational Modelling to Design for the Wellbeing Johanna Wald Dynamic 3D Scene Understanding

Dr. Christian Wachinger – LMU München: Computational Population Modeling from Big Medical Image Data (Nachwuchsforschungsgruppe) Prof. Dr. Björn Schuller – Universität Augsburg: Embedded Intelligence for Health Care and Wellbeing Prof. Dr. Oliver Amft – Universität Erlangen-Nürnberg: Digital Health Prof. Andreas Muxel – HAW Augsburg: Physical Human-Machine Interfaces

66 | SYMPOSIUM SYMPOSIUM | 67 SYMPOSIUM PROGRAMM

GRUPPENEINTEILUNG FÜR POSTERSESSIONS

SESSION B

Gruppe B1 (Raum „Alan Turing“) B1 Jakob Karolus Creating Proficiency-Aware Systems through Ubiquitous Sensing Technology Fiona Draxler It’s All Greek to Me – Support Language Learning with Ubiquitous Technologies Sarah Prange Supporting Users in Everyday Life Using Behavioural Biometrics Thomas Kosch Amplifying Cognitive Performance through Workload-Aware User Interfaces

Gruppe B2 (Raum „Margaret Hamilton“) B2 Lukas Mecke Informing users’ decision making towards the adoption of biometric techniques Andreas E. Schütz Sensitizing employees of small and medium-sized companies individually for information security Oliver Haas Digitalization in clinical environments using graph databases Henning Pridöhl On the influence of the Java Script Scripting Language and ist Ecosystems on Security and Privacy

Gruppe B3 (Raum „Konrad Zuse“) B3 Adrian Derungs Wearable sensor-based motion analysis and recovery trend quantification in hemiparetic patient in free-living Matthias Hoppe Haptic Feedback and Interaction in Virtual Reality Myroslav Bachynskyi Control-theoretical models for biomechanics-consistent control of arm movements in HCI Nina Rohrbach “Mixed Reality” as a new therapeutic approach to support activities of daily living in patients with chronic neurological disease

Gruppe B4 (Raum „Grace Hopper“) B4 Verena Süß Self-regulation in European Data Protection Law Tobias Pauli Inter-organizational business model interdependencies within the context of digitalization Bastian Haberer Empowering Users in the Data Economy Jürgen Hahn Sketchable Workspaces and Workflows

Gruppe B5 (Raum „Edsger Dijkstra“) B5 Michael Schmid Parallel & Concurrent Real-Time Design Patterns Barbara Eckl Human-Oriented Analysis of Software Developer Networks Tina Schuh Constructing a Semantic Map of Source Code Artifacts to Enhance the Software Development Process Alexander Wallis State-of-health estimation of lithium ion batteries

GRUPPENEINTEILUNG FÜR FELLOWS COACHING

C SESSION C D SESSION D E SESSION E Coaching C1: Christoph Molnar Coaching D1: Clemens Birkenmeier Coaching E1: Verena Süß (Raum „Alan Turing“) (Raum „Alan Turing“) (Raum „Alan Turing“) Coaching C2: Johanna Wald Coaching D2: Nina Rohrbach Coaching E2: Jonas Schlund (Raum „Margaret Hamilton“) (Raum „Margaret Hamilton“) (Raum „Margaret Hamilton“) Coaching C3: Sarah Wunderlich Coaching D3: Philipp Dufter Coaching E3: Alice Baird (Raum „Konrad Zuse“) (Raum „Konrad Zuse“) (Raum „Konrad Zuse“) Coaching C4: Oliver Haas (Raum „Grace Hopper“)

68 | SYMPOSIUM 69 SYMPOSIUM ABSTRACTS RELATED WORK

The following works have offered a foundation for aspects of this being, and reaffirming the need for a multimodal approach. Oord research project – acoustic based emotional analysis [4], multimod- et al. [6], present a state-of-the-art generative model for raw audio, al state analysis [5], and audio generation [6]. Weninger et al. [4], which is able to achieve high fidelity, from raw acoustic audio input. outline a standardised methodology for acoustic feature extraction This model itself will be a core consideration when developing the with corresponding emotion targets, showing that for general au- generation network and considering personalisation. dio the same methods can be applied to different audio streams With these factors in mind, the main contribution and novelty of producing similar results. These results offer a base to further de- this research project will be a novel combined analysis-synthesis Soundscape Generation for Wellbeing velop emotional state analysis from the acoustic domain. Carneiro approach i.e., multimodal monitoring and audio generation. More- et al. [5], present a method for monitoring stress in smart device over, improving on the state-of-the-art for synthetic audio genera- via Multimodal Individual Monitoring users, via mobile-ready sensors. Results from this study indicate tion, and acoustic monitoring will represent a strong outcome. 01 that monitoring multiple modalities can affectively measure well- Alice Baird PRELIMINARY RESULTS University of Augsburg, ZD.B Chair of Embedded Intelligence for Health Care and Wellbeing A number of publications have been accepted to date in relation to measurements, and real-time emotion annotation by three individ- [email protected] this project [4, 9, 10, 11, 12]. In particular, an ethical protocol was uals. The Emotional Synthetic audio (EmoSynth) dataset has also laid out for multimodal data acquisition and analysis in [7], which been generated, containing synthetic audio instances, labelled by Prof. Dr. Björn Schuller heavily informed the developed data acquisition protocols. Addi- 40 individuals for dimensions of valence and arousal. This data set tionally, ethical approval was obtained from the University of Augs- has been made public, and initial findings can be found in [9]. The burg Ethics Commission for further data acquisition. gathered BioS-DB dataset will be utilised for multimodal feature Following this, the BioSpeech Database (BioS-DB) of 55 individu- learning, amongst other machine learning tasks. Generative meth- INTRODUCTION als speaking aloud in front of others condition has been collect- ods to synthesize new instances of audio based on the provided ed, of which initial findings focusing on non-invasive monitoring labels of the EmoSynth dataset will be explored, and through trans- The notion of the soundscape was initially considered as an ecolog- this way, individual wellbeing is an ever present and challenging as- have been submitted for publication [8]. The BioS-DB consists of 2 fer learning methods, labels will be expanded to larger publically ical concept, in which culture and history can be captured through pect of modern life, and the impact of conditions including stress are channel audio, and 2 bio sensor signals, as well as self-assessment available datasets. sound, analogous to archival photography. When purposefully de- gaining prevalence [1]. With this in mind, this research project is pri- signed with synthetic audible combinations, the soundscape can marily exploring state-of-the-art computational methods to monitor augment an environment evoking specific emotional targets, able and improve states of wellbeing through the generation of personal- to enhance wellbeing, independent of cultural background [1, 2]. In ised synthetic listening experiences. PLANNED NEXT STEPS

RESEARCH QUESTIONS Next steps include the further analysis of the individual target group, will be explored include monitoring and adaptation of the synthet- with a focus on the efficacy of automated approaches for prediction ic environment based on bio-feedback. In this way explicitly gather of states of wellbeing from multiple modalities. Additionally, further qualitative evaluations from user experiences. In this same way, fur- The fundamental problem statement for this project is, how syn- Which methods for audio synthesis, and which combination of analysis of the synthesis methods, in the context of this user group ther analysis of the BioS-DB will be explored, with additional emo- thetically produced audio can be applied in the context of improv- audio can be used to improve wellbeing? (2) Which data modality will be explored. As initially research been a somewhat bottom-up tional annotations made offline, to make implicit commentary on the ing individual wellbeing. As well as this, can state-of-the-art deep and representation is most valuable as a means of monitoring well- approach, based on knowledge already gained a top-down analysis novel annotation approach implemented in this database. learning methods be applied for audio generation. Intelligent audio being? (3) Can reinforcement learning from continued self-assess- analysis (known as Computer Audition) and affective computing for ment be effectively applied to improve the target of audio genera- health and wellbeing are growing fields. In particular, deep learning tion? (4) Can raw audio generation, (e.g. via Generative Adversarial shows great promise for monitoring of health conditions via acous- Networks) be applied for realistic, high-quality and personalised OPEN ISSUES AND POSSIBILITIES FOR COOPERATION tic modalities, e.g. speech [2]. As well as this, urban populations listening experiences? (5) Can multimodal monitoring and audio are continuing to expand and without appropriate forethought, generation be realistically embedded for real-world use, e.g., within Currently a broad definition of wellbeing has been applied to the having feedback in regards to such deep learning methods would so too will noise pollution. Poorly considered sound environments smart-devices? end user, and the project may benefit from narrowing this down. be a benefit to the project at this stage. Within the ZD.B, this project can also have both long- and short-term implications for mental To this end, the expected contribution of this work would be three- Any feedback in relation to more specific users would be of value would benefit from any of the Digital Health initiatives, as well as dis- and physical wellbeing [3]. In this way, although currently focusing fold; audio synthesis methods to improve wellbeing, multimodal to the project. As well as this method for audio synthesis through cussion with members of groups such as the LMU, Human-Centred on individual monitoring, the outcomes of this project can be com- monitoring for states of wellbeing, and embedding of computation- raw modelling audio, may be a somewhat inefficient approach, and Ubiquitous Media Lab. bined for broader use-cases such as smart city infrastructure. ally expensive methods for audio analysis and generation in the Key research questions that this project will explore include: (1) real-world. REFERENCES

RESEARCH APPROACH AND METHODOLOGY [1] M. Jackson, "The stress of life: a modern complaint?", The Lancet, vol. 383, no. 9914, [8] A. Baird, S. Amiriparian, M. Berschneider, M. Schmitt and B. Schuller, "Predicting Blood pp. 300-301, 2014. Volume Pulse and Skin Conducatance from Speech: Introducing a Novel Database and [2] N. Cummins, A. Baird and B. Schuller, "The increasing impact of deep learning on Results (under review)", in Internation Conference on Acoustics and Speech Signal Process- This research project takes a pragmatic approach, with initial re- will aid data understanding, which will fine tune further quantitative speech analysis for health: Challeneges and Oppurtunities", Methods, Special Issue on ing (ICASSP), Brighton, 2019. search goals being mostly qualitative, leading to the extraction of ac- computational analysis. Final evaluation of this project is expected to Translational data analytics and health informatics, p. 25 pages, 2018. [9] A. Baird, E. Parada-Cabaleiro, C. Fraser, S. Hantke and S. Björn, "The Perceieved Emo- [3] L. Goines, "Noise Pollution: A Modern Plague", Southern Medical Journal, vol. 100, no. 1, tion of Isolated Synthetic Audio – A Dataset and Results", in Audio Mostly, Wrexham, quired data for more quantitative analysis. Given the varied aspects include a series of descriptive case and user studies, to demonstrate pp. 287-294, 2007. UK, 2018. of this project such an approach is vital, as qualitative observations the system efficacy in a variety of scenarios. [4] F. Weninger , F. Eyben , B. Schuller, M. Mortillaro and K. Scherer, "On the acoustics [10] P. Moscoso, M. Peck and A. Eldridge, "Systematic literature review on the association of emotion in audio: what speech, music, and sound have in common", Frontiers in between soundscape and ecological/human wellbeing", PeerJ Preprints, p. 28, 2018. Psychology, vol. 4, p. 292, 2013. [11] R. Cain, P. Jennings and J. Poxon, "The development and application of the emotional [5] D. Carneiro, J. Castillo, P. Novais, A. Fernadez-Caballero and J. Neves, "Multimodal dimensions of a soundscape", Elsevier, Applied Acoustics, vol. 74, no. 1, pp. 232-239, behavioral analysis for non-incasive stress detection", Expert Systems with Applications, 2013. vol. 39, no. 18, pp. 13376-13389, 2012. [12] A. Baird, E. Parada-Cabalerio, S. Hantke, F. Burkhardt, N. Cummins and B. Schuller, [6] A. v. d. Oord , S. Dieleman, H. Zen, K. Simonyan, O. Vinyals, A. Graves, N. Kalchbrenner, "The Perception and Analysis of the Likeability and Human Likeness of Synthesized A. Senior and K. Kavukcuoglu, "WaveNet: A Generative Model for Raw Audio", Arxiv, Speech", in Interspeech, Hyderabad, India, 2018. 2016. [13] A. Baird, S. Hasse Jorgensen, E. Parada-Cabaleiro, S. Hantke, N. Cummins and B. [7] A. Baird, S. Hantke and B. Schuller, "Responsible and Representative Multimodal Data Schuller, "Listener Perception of Vocal Traits in Synthesized Voices: Age Gender, and Aquisistion: On Auditability, Benchmarking, Confidence, Data-Reliance & Explainabili- Human-Likeness", Journal of the Audio Engineering Society, vol. 66, no. Special Issue, ty", in Language Resource Evaluartion (LREC), Miyzaki, Japan, 2018. p. 8, 2018

70 | SYMPOSIUM SYMPOSIUM | 71 ABSTRACTS form of a language specification of automotive industry stand- Grounded Theory to show that our approach is industry relevant SYMPOSIUM ards (namely AUTOSAR and EAST-ADL) in form of an annex. As and fits the needs for automotive software engineering experts soon as this annex is available, it is possible to model complex and security experts. Furthermore, we examined the related automotive software systems with a reasonable degree of se- work, e.g., the work of Holm [1], featuring a Cyber Security Mode- curity. For our first contribution we have identified the following ling Language (CySeMoL) for enterprise architectures, Mouratidis requirements for our language specification: [2] (Secure Tropos), papers such as “Modelbased security engi- • Classifying attacks and security threats. neering for cyber-physical systems: Asystematic mapping study” • Defining security goals. [3], featuring UMLSec which allows to express security- relevant • Extending the EAST-ADL metamodel with entities to represent information within the diagrams in a system specification, INCOSE actors and link them to consequences and affected modeling en- work on integrating system engineering with system security engi- IT-Security for Autonomous and tities. • Representing the attack vector and all its stages – from neering [4], NIST SP 800-160 and other NIST work on cyber-phys- attacker to breach – to affected vehicle features. ical systems. Connected Driving • A core solution idea for the attack vector handling. Our research 02 approach is backed by empirical validation of the results. We used Markus Zoppelt PRELIMINARY RESULTS

Nuremberg Institute of Technology, Faculty of Computer Science The first published result wasSAM : A Security Abstraction Model for By conducting a qualitative analysis in accordance with the method- [email protected] · zoppelt.net automotive software systems. It is a solution for modeling secure ology of Grounded Theory we gathered evidence that our solution Prof. Dr. Ramin Tavakoli Kolagari · Prof. Dr. Dominique Schröder automotive systems in the early system development phase in or- is relevant to the industry and is conform to the general paradigms der to reduce risks connected to late identification of security threats of automotive software engineering. By improving the identification and vulnerabilities in the automotive system life cycle. The approach and probing of security attack vectors, we provide a robust foun- tightly couples security management and model-based systems en- dation for automotive security testing. Afterwards, we start a bot- gineering by an abstract description of automotive security mode- tom-up approach to examine the basic software and communication INTRODUCTION ling principles. The resulting SAM language specification is based on layers to identify security flaws in modern vehicles and try to develop security requirements elicited from common industrial scenarios. It countermeasures, which are (cost-)effective. The current project is Due to the emergence of (semi-)autonomous vehicles and net- software-relevant quality objectives, in particular security, safety and is a suitable solution for representing attack vectors on vehicles and called “aID crypt” and enables basic encryption / masking of CAN-Bus worked technologies in the automotive domain, the development dependability. Moreover, the context of the industry is changing and provides a thorough security modeling for the automotive industry. messages to prevent certain attack vectors against vehicle networks. of secure and reliable vehicles plays an increasingly important role the controllability is shifting away from OEMs, i.e., they do not ful- The concept is developed and empirically proven through complete in the protection of road users. Safe and secure road transport is a ly control the aftermarket of external components such as smart- induction. major societal and political objective, which is substantiated by the phones, tablets, infotainment systems etc. Nonetheless, they are concrete goal of the European Commission to “move close to zero responsible for providing a common platform with secure interfaces fatalities in road transport" within the next three decades. to the required exterior components. Customers of automobiles ask for reliability, privacy and safety along with common convenience One historically often neglected aspect in automotive system devel- features like keyless entry and internet connectivity. In most cases, PLANNED NEXT STEPS opment is security-by-design, i.e., freedom from maliciously imple- these convenience features collide with basic security principles and mented threats. In the automotive software industry, model-based ultimately lead to a less secure product. A lack of security can effec- Future work will concentrate on implementing the results of our of automotive software and the wide range of cyber and physical engineering is the current state of the practice. Instead of integrat- tuate safety risks as well as endanger the driver, fellow passengers current work as a novel application for OEMs to apply and imple- attacks against it. Furthermore, our topdown approach for a func- ing security into the entire system development process, it currently and other road users. Car hacking, route sponsoring and even lethal ment security principles as presented in this paper. For example, tional architecture description provides the groundwork for one of tends to be an afterthought. Because of the tight interdependencies accidents may be the consequence. The manufacturer’s reputation is the concepts and benefits of SAM and our work could serve as a the main applications of networked vehicle technology: ECU archi- and integration of components, the consequences of gaping security at stake as well. Therefore, it is expedient to provide a platform-inde- motivation for the further development of the Adaptive AUTOSAR tectures for highly automated and autonomous driving. Another flaws are grave. pendent base for modeling secure software for automotive systems. Platform. Further work may also focus on developing an integrat- promising approach is the development of the “aID-crypt” encryp- ed development environment (IDE) for SAM that performs checks tion algorithm to enable authentication of CAN bus messages even The growing complexity of electrical automotive systems forces for consistency, completeness and integrity on each SAM-based on cheap hardware platforms without the necessity to upgrade original equipment manufacturers (OEMs) to have expertise in all software project, e.g., in MetaEdit+. Ideally, this would validate the existing hardware. scalability of SAM with respect to the size and real-world complexity PROBLEM STATEMENT AND RESEARCH QUESTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Developing secure automotive systems is a major challenge par- Secure automotive network architectures were not prioritized in ticularly for intelligent cars acting more as a computer than a tra- the past due to the general preconception in the last three decades One unresolved challenge concerning our project is the purchase CAN bus networks and ECU architectures. We need further external ditional vehicle. Modern cars are interconnected networks, with that cars are secure because of their technical complexity (security of a real-world vehicle to run testing, e.g., penetration tests, on. expertise in the area of lightweight cryptography and industry pro- potentially more than 150 Electronic Control Units (ECUs) in luxury by obscurity). However, numerous attack vectors on cars and their Some of the described attacks can only be sufficiently examined cesses of the automobile domain, e.g., of an OEM. models, communicating with each other and with the environment network of ECUs, actuators and sensors exist. In contrast to desk- when tried on a real vehicle. This is especially important for testing (Car-2-X communication). Attackers do not target cars in the same top computers, human lives are directly at stake when these “driv- way they would attack standard computer systems; cars use differ- ing computers” are the target of an attack. Cyber-attacks should ent networks, protocols and architectures. Countermeasures like always be considered highly critical when passengers’ lives are at firewalls do not exist for cars, yet. Moreover, cars carry- burden risk. Today, everybody is fully aware of the necessity to increase the REFERENCES some legacy mechanisms with insecure and unencrypted proto- security standard of cars and we are constantly reminded of that [1] Holm, H., Ekstedt, M., Sommestad, T., and Korman, M. A Manual for the Cyber Security [4] Ross, R., McEvilley, M., & Carrier Oren, J. (2016). Systems Security Engineering: Con- cols (e.g., CAN, Controller Area Network) in their system design and fact by regular press releases about attacks on cars. Modeling Language. siderations for a Multidisciplinary Approach in the Engineering of Trustworthy Secure were originally not designed in line with today’s security principles. [2] MOURATIDIS, H., & GIORGINI, P. (2007). Secure Tropos: a Security-Oriented Extension Systems, 160 (November 2016). https://doi.org/10.6028/NIST.SP.800-16 of the Tropos Methodology. International Journal of Software Engineering and Knowledge Engineering, 17(02), 285–309. https://doi.org/10.1142/S0218194007003240 [3] Nguyen, P. H., Ali, S., & Yue, T. (2017). Model-based security engineering for cyber-physical systems: A systematic mapping study. Information and Software Technol- RESEARCH APPROACH AND METHODOLOGY ogy. https://doi.org/10.1016/j.infsof.2016.11.004

We are using a top-down approach for developing secure and and classifying threats and risks and derive requirements from reliable automotive software systems. We start by identifying them. The described methods are captured and presented in the

72 | SYMPOSIUM SYMPOSIUM | 73 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

So far, a possible mechanism for flow induced coagulation in mem- ity in complex geometries were discussed with various experts (a brane oxygenators was identified (these preliminary results were conference talk was given with subsequent extensive discussion). presented as talks at two conferences and a manuscript is in review). To prove the proposed coagulation mechanism, a suitable An important but so far missing finding would be the dependen- experiment was developed. The feasibility of retrieving relevant cy of the previously described mechanism on patient’s condition A realistic and patient specific information by using an improved computed tomography imaging and medication. An important result would be to bring it down to a technique was demonstrated (a short-communication was pub- small set of clinical relevant parameters. computational model of blood viscosity lished, a poster was presented and a manuscript is currently in 03 and flow induced coagulation preparation). Two possible modelling approaches for blood viscos- Clemens Birkenmaier PLANNED NEXT STEPS A next step is to perform a confirmatory experiment on the flow in- However, the information gained from several experiments and University Regensburg, Faculty of Medicine · OTH Regensburg, Faculty of Mechanical Engineering duced mechanical coagulation pathway and identify model param- from clinical data needs to be put into computational models. In [email protected] eters. In addition, this process will be simulated using micro-scale order to tie up the loose ends, a successful integration into a single Prof. Dr. Karla Lehle · Prof. Dr. Lars Krenkel mechanics computations. To be able to parametrise and validate computational framework is necessary. the viscosity modelling approaches, also suitable experiments need to be performed. An important step will be the identification of a parameter set, describing the relevant dependencies and at INTRODUCTION the same time being part of the clinical every day routine.

Complex circulatory interventions, such as extra corporeal mem- knowledge of the interaction of blood flow, artificial device and pa- brane oxygenation, can so far only be a rescue therapy in critically tient specifics. To make this broad field and the underlying complex OPEN ISSUES AND POSSIBILITIES FOR COOPERATION ill patients and bear a high risk of complications [1]. Despite all im- phenomena more accessible for physicians making therapy deci- provements and progress, the understanding of blood behaviour in sions as well as engineers developing artificial organs, an efficient The automated image analysis of immunofluorescence micrographs framework. In this field, expertise from the ZD.B crowd would be re- any complex artificial device (e.g. artificial heart or lungs) lacks the computational model is necessary. turned out to be difficult, especially when samples are compromised ally appreciated. Taking into account the enormous number of rele- or just sparsely populated. Effective analysis of fluorescence micros- vant parameters, providing effective decision support for medication copy images can be a challenge, where cooperation could be really and operating parameters of extracorporeal life support systems is PROBLEM STATEMENT AND RESEARCH QUESTION beneficial. Another topic to be potentially difficult is to implement a huge challenge. As there is missing a deterministic (or even a heu- a satisfying transition from deterministic to statistical model behav- ristic) modelling approach, an artificial intelligence based prediction iour due to a lack of information or resolution. A further challenge system could bridge this gap. In this field, the knowledge of the ZD.B Simulation based individual modelling and prediction is becoming model will facilitate a better outcome in extracorporeal life support might become linking the different modelling approaches (finite-vol- network could help a lot. increasingly important in the so-called precision medicine. Addi- therapy. This project establishes the so far missing link between the ume, lattice-Boltzmann, lumped parameter) into one single powerful tionally, digital prototyping in industry is already extensively used in highly nonlinear viscosity of blood and its mechanical coagulation product development. Both need reliable and economically appli- pathway. Besides an improved understanding of nonlinear fluids in cable tools. By answering the question why blood is clotting in artificial general and methodological progress in efficient numerical meth- lungs and by developing a patient specific computational model, a step ods, the findings of this project will help improving the survival rate REFERENCES towards simulation based decision making in therapy is made as of critically ill patients by potentially affecting clinical guidelines and well as a key tool for reliable digital prototyping of a new generation pave the way for more reliable medical devices. [1] Lubnow M, Philipp A, Foltan M, Enger TB, Lunz D, Bein T, et al. Technical complications [3] Casa LDC, Deaton DH, Ku DN. Role of high shear rate in thrombosis. Journal of vascu- during veno-venous extracorporeal membrane oxygenation and their relevance predict- lar surgery. 2015 Apr;61(4):1068–1080. of extracorporeal life support devices is provided. Providing such a ing [4] Závodszky G, Rooij B van, Azizi V, Hoekstra A. Cellular Level Modeling of Blood Rhe- a system-exchange–retrospective analysis of 265 cases. PloS one. 2014;9(12):e112316. ology with An Improved Material Model for Red Blood Cells. Frontiers in physiology. [2] Dornia C, Philipp A, Bauer S, Hoffstetter P, Lehle K, Schmid C, et al. Visualization of 2017;8:563. thrombotic deposits in extracorporeal membrane oxygenation devices using multide- [5] Fedosov DA, Noguchi H, Gompper G. Multiscale modeling of blood flow: from single RESEARCH APPROACH AND METHODOLOGY tector computed tomography: a feasibility study. ASAIO Journal. 2013;59(4):439–441. cells to blood rheology. Biomech Model Mechanobiol. 2014;13(2):239–258.

The conducted work is based on the hypothesis that there is a blood relevant clinical data and high-resolution computed tomography im- coagulation pathway, which is triggered or at least mediated by flow aging of clinically used oxygenator devices, are implemented. and besides that only dependent on some parameters, describing patient’s condition and medication. So far, the exact mechanism be- Using computed tomography imaging in order to get relevant infor- hind the observed phenomena remained unclear. mation on coagulation status inside membrane oxygenators, Dornia [2] has been the first to show its feasibility. In the present project, the At the outset, the clot formation and progression mechanism in mem- obviously fundamental lack of imaging resolution is uncovered and brane oxygenators is investigated in detail using histological methods an improved method will be proposed. The Ku group [3] published and image analysis tools. From this, a model for the flow-mediated outstanding research on mechanically triggered coagulation over pathway is drawn. A second step comprises of finding and introduc- the past decade. However, it seems that they are missing a relevant ing a suitable numerical viscosity representation, circumventing the connection, which links their findings to events reported in artificial shortcomings (accuracy, efficiency or scalability) of present models. lungs. The groups of Hoekstra [4] and Fedosov [5] have put much ef- For evaluation purposes, micro particle image velocimetry and parti- fort in computational modelling of blood behaviour in depth and de- cle tracking experiments are used, which requires proper determin- tail. Bringing this accuracy in detail on a more global level makes an istic and statistical image analysis on several levels in time and space. application to a relevant scenario feasible. This project aims on small A third step consists of combining the two aforementioned models enhancements in several fields, but on a considerable step forward into a suitable computational framework in a scalable and efficient in understanding and predicting the phenomenological relevant pro- manner. This is crucial, since the actual application comprises huge cesses, and at the end of the day, making patents’ lives longer (or at cases. Additionally, patient specifics, found by statistical analysis of least a little less worse).

74 | SYMPOSIUM SYMPOSIUM | 75 ABSTRACTS The second research topic focusses on the evaluation of net- Furthermore, a new workflow for detecting malicious activities SYMPOSIUM work-based intrusion detection methods. Labelled intrusion de- within flow-based network traffic has been developed [3]. The tection data sets seem to be a valuable approach for evaluation. main idea of this workflow is to enrich the flow-based data stream However, existing labelled data sets are either outdated or can- with additional information. The additional information is directly not be published due to privacy reasons. Furthermore, the use of integrated in the analysis algorithms or is provided as an addi- real network traffic is also problematic, since there are no labels tional data source. In contrast to previous work, our workflow [3] (ground truth) for it and attack scenarios cannot be tested in com- enriches the data stream through aggregations over sliding time pany networks. Therefore, the second research topic investigates windows, includes knowledge of domain experts and evaluates the generation of labelled flow-based network data. For this pur- the flow-based data from three different views. Incorporating do- pose, the network of a small business company was rebuilt in a main knowledge and aggregating flows over time windows allows Data Mining for Network-based virtual environment and normal user behaviour is emulated with for the calculation of more meaningful features for subsequent parameterized Python scripts. Various attack scenarios may be analysis algorithms. Overall, this workflow should improve the de- Intrusion Detection executed in this environment. Since automated scripts are used tection of novel attacks while simultaneously reducing the num- to control normal user behaviour and attack scenarios, exact la- ber of false alarms. 04 belling of the captured network traffic is feasible.

Markus Ring PRELIMINARY RESULTS Coburg University of Applied Sciences and Arts [email protected] Interim results are achieved in all three research areas. For han- data sets are publicly available and contain normal user behaviour Prof. Dr. Andreas Hotho · Prof. Dr. Dieter Landes dling categorical attributes, two similarity measures ConDist [4, 5] as well as different attack scenarios. and IP2Vec [6] have been developed. While ConDist is a general similarity measure for heterogeneous data, IP2Vec is focused on Attackers invariably utilize port scans to identify target hosts and flow-based network traffic. IP2Vec [6] was evaluated on twosce- networks. Consequently, port scans are often indicators of more narios: client-server identification and botnet identification. For both serious attack scenarios. The proposed network-based anomaly INTRODUCTION scenarios, initial results are very promising. detection workflow [3] was applied for detecting (slow) port scans with good results [9]. Cyber attacks against small- and medium-sized companies (SMEs) can neither detect, nor prevent so-called insider attacks and ze- The virtual test environment was used to create the labelled CI- are frequent these days. The increasing digitalization through In- ro-day exploits. The crypto trojans Locky and WannaCry or the heart- DDS-001 [7] and CIDDS-002 [8] intrusion detection data sets. Both dustry 4.0 and other developments exposes SMEs to new threat bleed bug in OpenSSH are well-known examples of zero-day exploits scenarios. Simultaneously, studies indicate that SMEs are becom- that have caused considerable damage to companies. The disser- ing increasingly aware of cybercrime. Companies usually use vari- tation project "Data Mining for Network-based Intrusion Detection" ous security mechanisms like firewalls, virus scanners or intrusion investigates the suitability of different data mining methods for at- detection systems (IDS) to protect their confidential data. However, tack and insider threat detection within company networks. PLANNED NEXT STEPS such systems only protect against known attacks from outside and Further steps include the creation of an additional intrusion detec- the proposed port scan detection approach [9] was successfully used tion data set with improved normal user behaviour as well as more on attacks in their scanning phase, further research effort needs to PROBLEM STATEMENT AND RESEARCH QUESTION sophisticated attack scenarios. Furthermore, we intend to use deep be spent on detecting attacks in the gaining access phase. neural networks for the analysis of flow-based network traffic. While Data mining algorithms are successfully applied in various appli- A major technical challenge is that flow-based network data de- cation domains. However, the breakthrough in IT security is still scribe network connections (no payload information) and encom- missing due to the special challenges (e.g. missing publicly avail- pass continuous and categorical attributes. However, basic math- OPEN ISSUES AND POSSIBILITIES FOR COOPERATION able data sets or the high cost of false alarms) of this domain [1]. ematical operations used in standard data mining algorithms can Additionally, new security systems for SMEs need to easily integrate only be applied to continuous attributes since categorical attributes Realistic user simulation is an ongoing research question. Further users. Feedback on modelling user behaviour would be helpful and (preferably without additional hardware) into existing IT infrastruc- have no natural ordering. One question arises in this context which improvements can be made by modelling the exact behaviour of real could be the basis for possible cooperations. tures and simultaneously respect the privacy of employees. There- is part of the dissertation: (2) How to use available information of fore, this dissertation focusses on flow-based network traffic which categorical attributes in standard data mining algorithms? There- can be easily captured at network components like firewalls and fore, the dissertation analysis the calculation of meaningful distanc- routers without compromising the privacy of employees. es between categorical attributes and the transformation of cate- REFERENCES gorical attributes into numerical spaces Rn while preserving their [1] R. Sommer, P. Vern: "Outside the Closed World: On Using Machine Learning for Net- [6] M. Ring, A. Dallmann, D. Landes, A. Hotho: “IP2Vec: Learning Similarities between IP Company networks have different usage profiles depending on day- semantics. work Intrusion Detection", IEEE Symposium on Security and Privacy (SP), IEEE, 2010, pp Addresses,” in Workshop on Data Mining for Cyber Security (DMCS), International to-day business, and therefore the observed data stream changes 305-316. Conference on Data Mining (ICDM), IEEE, 2017, pp. 657-666. continuously. Novel attacks and threats can only be detected if nor- In addition, the dissertation addresses the evaluation of analysis [2] T. Mikolov, K. Chen, G. Corrado, J. Dean: “Efficient Estimation of Word Representations [7] M. Ring, S. Wunderlich, D. Grüdl, D. Landes, A. Hotho: “Flow-based Benchmark Data in Vector Space,” in arXiv preprint arXiv:1301.3781, 2013. Sets for Intrusion Detection,” in European Conference on Cyber Warfare and Security mal user behaviour is known for company networks. The challenge results, giving rise to another research question: (3) How to eval- [3] M. Ring, S. Wunderlich, D. Grüdl, D. Landes, A. Hotho, “A Toolset for Intrusion and (ECCWS), ACPI, 2017, pp. 361–369. is to distinguish the change in the data stream (concept drift) from uate the quality of network-based intrusion detection methods in Insider Threat Detection,” in Data Analytics and Decision Support for Cybersecurity: [8] M. Ring, S. Wunderlich, D. Grüdl, D. Landes, A. Hotho: “Creation of Flow-Based Data malicious network activities. Therefore, one research question of general? Trends, Methodologies and Applications, I. Palomares, H. Kalutarage, and Y. Huang, Sets for Intrusion Detection,” in Journal of Information Warfare (JIW), Vol. 16, Issue 4, Eds. Springer, 2017, pp. 3–31. 2017, pp. the dissertation is: (1) How to model normal user behaviour for [4] M. Ring, F. Otto, M. Becker, T. Niebler, D. Landes, A. Hotho: “ConDist: A Context-driven 40-53. company networks? Categorical Distance Measure,” in European Conference on Machine Learning and [9] M. Ring, D. Landes, A. Hotho: “Detection of Slow Port Scans in Flow-based Network Knowledge Discovery in Databases (ECML-PKDD). Springer, 2015, pp. 251–266. Traffic”. In PLoS ONE, Vol. 13, Issue 9, 2018, pp. 1-18. [5] M. Ring, D. Landes, A. Hotho: “Automatic Threshold Calculation for the Categorical Distance Measure ConDist,” in LWA 2015 Workshops: KDML, FGWM, IR, and FGDB, RESEARCH APPROACH AND METHODOLOGY 2015, pp. 52–63.

The dissertation has three major research topics. The first re- and uses a neural network to transform categorical values into a search topic emphasises the handling of categorical attributes. numerical space Rn. The other approach uses conditional prob- Therefore, this research investigates how available context inability distributions of correlated context attributes in order to formation can be used to derive similarities between categorical create a similarity matrix between categorical values. values. One approach is inspired by text mining (Word2Vec [2])

76 | SYMPOSIUM SYMPOSIUM | 77 SYMPOSIUM ABSTRACTS RELATED WORK

Considering a blockchain based organization of distributed assets chains suggesting possible benchmarks is described in [8]. They con- the first project was a local energy market (LEM) called The Brooklyn clude that there is still a big performance gap between blockchains Microgrid [4]. This pilot was followed by research projects like the and current databases. LAMP project in Germany. In [5] first ideas for blockchain design rec- Most of related work aims for showing or implementing possible use ommendations are provided. Possible current blockchain solutions cases. This PhD-project intends to analyze existing and think of new considering energy transactions are summarized in [6]. use cases, to develop an unbiased evaluation framework for such In terms of evaluation metrics considering blockchain technology a use cases, to identify the most suitable DLT designs and to quantify Blockchain-Based Orchestration simple flow chart is offered in [7]. However, it is unquantified, not the extra effort of a DLT-based approach in comparison to central- sufficiently differentiated and only applied for use cases outside the ized approaches for the most promising use cases. Therefore, it com- 05 in Decentralized Power Systems energy sector. A first general evaluation framework for private block- plements and fits in well in the overall picture. PRELIMINARY RESULTS

Jonas Schlund RQ 1 is partially answered and first ideas considering RQs 2 and 3 coordination and sharing of charging stations. This is chosen as it is are summarized in the following. RQ 4 builds upon the previous expected that the share of BEVs will increase drastically in the next Friedrich-Alexander-University Erlangen-Nürnberg RQs and can thus not yet be tackled. years and because uncoordinated charging might cause problems [email protected] in distribution grids. A trusted third party is undesirable as there Prof. Dr. Reinhard German RQ 1: Considering RQ 1 the use cases of interest have been identi- are many different charging station operators and energy provid- fied. For each use case an always online trusted third party is un- ers involved. As of today, problems already exists considering the desirable or disadvantageous, it is necessary to store state and to interoperability between different providers. allow for multiple and possibly untrusted or unknown writers of INTRODUCTION data. The identified use cases are clustered into three subgroups. RQ 2: In terms of RQ 2 ideas for the metrics have been determined, The first subgroup is P2P electricity or certificate trading on differ- however the formal definition of the metrics is still pending. Two Our society depends on a reliable power supply. Therefore, the elec- In this context many recent studies and publications attribute a high ent levels, from device level over local energy markets and com- different classes of metrics have been identified: quantitative met- trical power system (EPS) is traditionally planned and operated hierar- potential to distributed ledger technologies (DLTs) [1-3]. Therefore, I munities up to trades between transmission system operators. A rics, which can be derived by means of the simulations and quali- chically and centrally. However, this does not reflect the increasingly investigate if and how an orchestration of assets in the EPS is possi- central third party contradicts the idea of P2P trading and obviously tative metrics, which only play a secondary role in the PhD project. decentralized and volatile energy landscape. The rising number of ble based on distributed peer-to-peer (P2P) IT such as blockchains or storing state is necessary for being able to trade. In addition, a trad- The quantitative metrics are the general performance from the highly heterogeneous actors in the EPS results in an increased total other DLTs. ing partner might be unknown and therefore possibly untrusted. A point of view of the EPS, the expenses from the point of view of complexity of the system. Many of the newly installed distributed as- prototypical implementation of a local energy market with focus on the IT system, the scalability and the level of decentralization. In sets generating renewable electricity are owned by citizens or small grid friendliness is currently in work. total, the cost of disintermediation is valuated by restraints in gen- companies and are thus not controlled centrally. How can such distrib- The second subgroup is the provision of flexibility as an ancillary eral performance, the additional holistic costs and the limitations uted assets be orchestrated in future? service for the EPS. In comparison to the market based approaches in scalability. above this approach faces the challenges from the problem statement in a more straightforward way. In this context DLTs have the RQ 3: Considering the DLT design first implications can be derived potential of organizing and mobilizing small scale distributed flex- from the gained experience from the literature review and the dif- PROBLEM STATEMENT AND RESEARCH QUESTION ibility sources, which do not have access to a flexibility market so ferent lab setups. At the current stage of the project permissioned far. Possible applications are frequency stability, voltage stability consortial blockchains or hierarchical hybrid public and consor- The underlying problem is that with a higher share of renewable en- trated in a distributed way. In particular: how well perform DLT-based or congestion management. A first prototype of our new concept tial proof-of-authority or practical byzantine fault tolerance mul- ergies more flexibility is needed in the EPS. Currently the increasing approaches in this context in comparison to centralized approaches? In of a self-organized virtual power plant is already implemented and ti-chains seem promising for most use cases. Alternatives based on number of decentralized assets are often not orchestrated and thus order to answer this question the central RQ is split into the following running in the lab, a pre-stage of it is published in [9]. directed acyclic graphs are still very immature but promise a better their possible flexibility in operation is not utilized. Without solving subquestions: The last identified use case is battery electric vehicle (BEV) charging scalability. this problem, the EPS is in future either going to be less reliable or 1. Which are the use cases of interest? over-dimensioned and more expensive from a macroeconomical 2. Which metrics are of importance in order to quantify how well point of view. DLT-based approaches suit the selected use cases? How are these PLANNED NEXT STEPS metrics formally defined and quantified? Assuming the hypothesis that a distributed approach is desired in 3. Which possible different DLT designs are most suitable and how The next steps are the formal definition of the evaluation frame- simulation models can be developed, the case studies can be eval- order to (self-)coordinate distributed assets in the EPS for political do different designs influence the determined metrics? work and a subsequent analysis of the influence of different block- uated and recommendations can be formulated. or philosophical reasons or owner preferences, the central research 4. How well perform and how large is the extra effort of DLT-based chain designs on the metrics. Once these steps are completed question (RQ) of this project is (how) can distributed assets in a fu- approaches in comparison to a centralized approach for the se- ture EPS empowered by renewable energies and storages be orches- lected use cases from a holistic point of view? OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

RESEARCH APPROACH AND METHODOLOGY So far a field trial with real assets is not yet planned. A cooperation ject. Alternatively, a cooperation with an existing field trial with the considering a new field trial would be beneficial for the research pro- possibility to use data from the field is desirable. The general idea of the PhD-project is to simulate the use cases of in- tative part prototypical implementations and experimental test runs terest with different DLT networks in order to evaluate them accord- are used for data collection and extensive simulation and modeling ing to a defined evaluation framework and to use prototypical imple- is used for studying the use cases. In order to mind both IT and EPS REFERENCES mentations for the validation of the simulation models. Therefore, layer, hybrid simulation including discrete and continuous simula-

the proposed methodology consists of a theoretical and an empirical tion will be used. [1] Bundesverband Energie und Wasserwirtschaft, Blockchain in der Energiewirtschaft, 2017. [6] G. Zizzo, E. R. Sanseverino, M. G. Ippolito, M. L. D. Silvestre und P. Gallo, „A Technical part. The theoretical part includes literature reviews, an analysis of [2] M. Merz, „Potential of the Blockchain Technology in energy trading,“ in Blockchain Approach to P2P Energy Transactions in Microgrids,“ IEEE Transactions on Industrial current research developments and analytical work. In the quanti- technology introduction for business and IT managers, de Gruyter, 2016. Informatics, pp. 1-1, 2018. [3] V. Schlatt, A. Schweizer, N. Urbach and G. Fridgen, Blockchain: Grundlagen, Anwendun- [7] K. Wüst und A. Gervais, Do you need a Blockchain?, 2017. gen und Potenziale, Fraunhofer FIT, 2016. [8] T. T. A. Dinh, R. Liu, M. Zhang, G. Chen, B. C. Ooi und J. Wang, „Untangling Blockchain: [4] E. Mengelkamp, J. Gärttner, K. Rock, S. Kessler, L. Orsini und C. Weinhardt, „Designing A Data Processing View of Blockchain Systems,“ IEEE Transactions on Knowledge and microgrid energy markets: A case study: The Brooklyn Microgrid,“ Applied Energy, Bd. Data Engineering, Bd. 30, pp. 1366-1385, 7 2018. 210, pp. 870-880, 2018. [9] J. Schlund, L. Ammon und R. German, „ETHome: Open-source Blockchain Based Ener- [5] D. Vangulick, B. Cornélusse and D. Ernst, "Blockchain for peer-to-peer energy exchang- gy Community Controller,“ in Proceedings of the Ninth International Conference on Future es: design and recommendations," in Proceedings of the XX Power Systems Computation Energy Systems, New York, NY, USA, 2018. Conference (PSCC2018), 2018. 78 | SYMPOSIUM SYMPOSIUM | 79 ABSTRACTS concerned with the mathematical foundation and the software im- approach is the creation of an interpretable machine learning model SYMPOSIUM plementation. These groups usually focus more on the psychological for a specific application. We don’t focus on a specific use case, but side of explanation systems (human-machine interaction). Another build a general toolbox that should be broadly applicable.

RELATED WORK

Work on an explanation method for individual predictions by Ribei- mAchine Learning EXplanations), a software project similar to the iml ro et al. [4] sparked my personal interest in interpretable machine toolbox we are implementing. Compared to DALEX, the iml package learning. The authors proposed a very general explanation method focuses more on model-agnostic methods and implements all meth- that explains the complex prediction model with interpretable, in- ods itself rather than relying on other packages. This has the advan- Interpretable Machine Learning dividual models for each prediction. The idea behind the method is tage of providing a more consistent interface and better control. fascinating and has been cited very often. However, it proved difficult 06 to apply their method in practice as it has hidden parameters that Guidotti et al. [6] have conducted an extensive survey of black box should be adapted to each new situation and it is unclear how these interpretability methods. While the paper covers many methods not parameters can be set automatically. We plan to create simpler in- covered in the Interpretable Machine Learning book [1], I believe that Christoph Molnar terpretability methods that are more robust and easier to check for an extensive book covering the basic ideas with application exam- Poster correctness. ples and critical discussions is a better format for teaching and pro- LMU Munich, Computational Statistics Award moting interpretable machine learning. [email protected] 3. PREIS Other important work is the software toolbox DALEX [5] (Descriptive http://www.compstat.statistik.uni-muenchen.de/people/molnar/ 500€ Prof. Dr. Bernd Bischl PRELIMINARY RESULTS

INTRODUCTION We have created the iml R package, which implements a wide range I have covered many methods in the book (still in progress), which to of methods for machine learning interpretability and has been the best of my knowledge is the most comprehensive source on the Machine learning is a set of techniques that gives computers the tions, as the computer can pick up associations in data that a hu- downloaded 3200 times so far (as of 01.09.18). We currently improve topic of interpretable machine learning with thousands of readers. ability to learn to make predictions from data. Machine learning man can’t. This superiority in predictive performance comes with the speed of the software and add more interpretability methods. is increasingly used in many products, processes and research. In a disadvantage in the form of loss of understanding the learned We expect to publish research on explanation methods for individual many applications, machine learning is superior to human predic- programs. predictions.

PROBLEM STATEMENT AND RESEARCH QUESTION

When computers learn by themselves, we no longer know how they machine learning model. PLANNED NEXT STEPS make decisions. Prediction models can become very complex, for What information can we reliably extract from a black box machine example a neural network consists of millions of weight numbers or learning model if we can only manipulate its inputs and observe the I plan to finish the first version of the book on “Interpretable Machine To complete my dissertation I need at least one or better two me- a random forest consists of hundreds of deep decision trees. This is predictions but have no insights into the model parameters? What Learning” this year. Furthermore, I plan to improve and extend the thodical first-author papers, in which I either develop / extend an problematic because we need transparency to ensure the safety of information about a black box model helps to better understand its iml interpretability toolbox based on own ideas and user feedback. interpretability method or test other methods extensively. We are the software, to detect possible biases, to foster trust, to debug the behaviour? What are limitations or (sometimes hidden) assumptions High priority is given to improving the speed and memory consump- currently exploring simple counterfactual explanations to explain in- system if something goes wrong and more. These problems make for existing interpretability tools and when are their results reliable? tion of the package. I also plan to add more model-agnostic inter- dividual predictions of a machine learning model, which can also be many companies and researchers reluctant to apply machine learn- How can we decide whether one interpretation method is better pretability methods to the package. used to reveal insights about the global model behaviour. ing. than another? All these research questions relate to the (very difficult) question of how we define or measure interpretability at all. In the field of interpretable machine learning, we deal with the problem of making complex machine learning models interpretable, By exploring interpretable machine learning, we want to mature the OPEN ISSUES AND POSSIBILITIES FOR COOPERATION meaning we develop tools to describe the overall prediction model field and increase the availability of reliable and well understood and to explain individual predictions. This includes both the meth- methods to explain the behaviour of machine learning models. The The (very difficult) question how we define interpretability remains I am also interested in industry or academia partners who already odological development of the tools, as well as their implementation availability of interpretability tools will hopefully catalyse the accept- unresolved. This impacts my work because it is difficult to evaluate use machine learning and would like to have an interpretability tool in software. In our group, we are especially interested in methods ance of machine learning in research and practice by increasing the new methods due to the lack of definition. Ideas on how to evaluate for the predictions and model behaviour. It would be great to discuss that are model-agnostic, i.e. they work with any type of supervised interpretability of computer decisions. explanations of predictions or other summary statistics of machine current needs and get feedback of users of the iml package. In par- learning models would be very helpful. ticular, it would be helpful to know which interpretability methods are most needed by users from industry and science. RESEARCH APPROACH AND METHODOLOGY

In my research, I consolidate knowledge in the form of a book on in- model-agnostic machine learning interpretability, then implement- REFERENCES terpretable machine learning [1], create an interpretability software ing them in software and testing them in different scenarios (i.e. [1] Molnar, C. (2018). Interpretable Machine Learning. Retrieved from [5] Biecek P (2018). “DALEX: explainers for complex predictive models.”_ArXiv e-prints_. toolbox [2] and develop methods. different types of machine learning models and different datasets). https://christophm.github.io/interpretable-ml-book/ 1806.08915, https://arxiv.org/abs/1806.08915 With this approach we want to ensure the usability and quality of [2] Molnar et al., (2018). iml: An R package for Interpretable Machine Learning. [6] Guidotti, Riccardo, et al. "A survey of methods for explaining black box models." arXiv The field of interpretable machine learning is very young and the lack the interpretability methods. In addition, we use benchmarks to Journal of OpenSource Software, 3(26), 786, https://doi.org/10.21105/joss.00786 preprint arXiv: 1802.01933 (2018). https://arxiv.org/abs/1802.01933 [3] R Core Team (2017). R: A language and environment for statistical computing. of extensive secondary educational literature has motivated me to measure the quality of the methods, for example how sparse an ex- R Foundation for Statistical Computing, Vienna, Austria. consolidate and thoroughly examine existing research in the form of planation is. We also simulate datasets and models where we know URL https://www.R-project.org/. a book for practitioners. the true explanations so that we can test our methods against this [4] Ribeiro et. al (2016). Why should i trust you?: Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD international conference on knowledge ground truth. discovery and data mining. ACM. In our group we develop new interpretability methods and analyse existing methods in greater depth. We conduct our research by de- Other research institutes test the quality of interpretability methods veloping or improving conceptual and mathematical methods for with surveys or experiments with human participants. We are more

80 | SYMPOSIUM SYMPOSIUM | 81 ABSTRACTS attacks and whether there are any attacks possible which have not hicle can look. Lin et al. do not cover all use cases but mostly focus SYMPOSIUM yet been reviewed. Last but not least, sufficient counter-measures on object detection and tracking, path and way finding and decision have to be developed. Both, the attacks and countermeasures, will making. They do not discuss system surveillance and machine learn- be implemented and tested on a real car. ing focused on the environment or entertainment, etc. There have been attempts at structuring published attacks by Bar- To gain deeper knowledge about security for machine learning in reno et al. [1] and Papernot [2]. Papernot has published many inter- autonomous driving, the above mentioned papers have to be fur- esting papers about attacks on machine learning systems during the ther discussed and missing de-tails have to be filled in. All possible last couple of years. use cases of ma-chine learning in autonomous cars have to be eval- Another promising paper was published by Lin et al. [3] It shows an ua-ted and their security standard has to be discussed. Security for Machine Learning in the idea of how a machine learning architecture in an autonomous ve- 07 Context of Autonomous Vehicles PRELIMINARY RESULTS After doing some research, it was soon clear that security in machine • Attacks that tamper with the training data or the training algorithm. learning seems to be not yet well known among machine learning • Attacks that change the model‘s code after deployment, e.g., re- Katja Auernhammer engineers. However, there is a variety of published attacks on ma- training the model. chine learning, which can be divided into six main classes: • Attacks that do not attack the machine learning model itself but the Nuremberg Institute of Technology • Attacks conducted by sending quantitative inputs to the system to surrounding system, e.g., the components that preprocesses data. [email protected] · https://th-nuernberg.de/nc/person/auernhammer-katja/ shut it down, e.g., to cause denial of service. When there will be further attacks published, the categories might Prof. Dr. Ramin Tavakoli Kolagari · Prof. Dr. Felix Freiling • Attacks using perturbed inputs to cause faulty outputs. undergo some changes in the future, e.g., defining the categories by • Stealing the machine learning system by extracting information adding more detailed subcategories. about the training data or model itself. INTRODUCTION

Autonomous driving is a field of active research. Not only scientists security for machine learning to keep the passengers and other road PLANNED NEXT STEPS and OEMs (original equipment manufac-turers) are dedicated to the users safe. development of autono-mous vehicles but also companies outside Since OEMs do not publish the current state of the art of machine After defining the categories, the algorithms and at-tacks will be To increase the robustness and security of machine learning sys- the field, e.g., Google and Apple are interested in positioning them- learning in cars, researchers face some difficult to overcome obsta- mapped to autonomous vehicles to create some form of road map tems, humans need to understand how these systems work. A selves on the growing market. OEMs claim that fully-autonomous cles. of attacks and countermeasu-res for machine learning in cars. „whiteboxing“ of such systems is necessary. However, such white- cars will be implemented by 2025. Although some scientists are not To conduct meaningful research of security for machi-ne learning in It is especially important to discuss if there are attacks that might boxing might also have implications on security: If developers can yet sure if this level of autonomous driving is applicable, autono- autonomous cars, use cases for machine learning in a car have to be not be applicable to the area of autonomous driving due to restric- look into their machine learning system, attackers might be able to mous driving will still be a field of high interest by that time. identified and mapped to appropriate machine learning algorithms. tions caused by the hardware or software architecture of an em- do that as well. An essential part of autonomous driving is machine learning used Based on these assumptions, published attacks on machine lear- bedded system. On the other hand, the characteristics of embed- in different use cases, e.g., object dete-ction and tracking, decision ning have to be evaluated, categorized and transfered to the field of ded systems could enable as-yet unkown forms of attacks. making, system surveillan-ce or entertainment of passengers. It is autonomous driving. Last but not least, countermeasures have to be important to realize that machine learning components are not im- defined and implemented. mune to attacks by hacker organizations with bad intentions. Since human life is at stake, it is crucial to conduct research in the field of OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

Insight knowledge of machine learning algorithms used in autono- Furthermore it is important to not only focus on algo-rithms cur- mous vehicles is crucial to conduct worthwhile research in the area. rently used in the area, but also to look ahead into the future. Only PROBLEM STATEMENT AND RESEARCH QUESTION Cooperation with an OEM would be quite helpful to conduct mean- Machine learning specialists might know which machine learning al- ingful research. gorithms might be relevant in the future. So far, there has been done quite some research in the fields of se- • What attacks are successful against machine learning systems in curity for machine learning and security for vehicles. However, no cars and what are the consequences in a worst case scenario? known research was conducted in the combined field, namely se- • How can machine learning in cars be secured against such and fu- curity for machine learning in autonomous vehicles. This combined ture attacks? REFERENCES field becomes increasingly important the further develop-ment of • Is it possible to apply some form of comprehensibility to machine [1] Barreno, Marco; Nelson, Blaine; Sears, Russell; Joseph, Anthony D.; Tygar, J. Doug. Can [3] Lin, Shih-chieh; Zhang, Yunqi; Hsu, Chang-hong; Skach, Matt; Haque, Md E.; Tang, autonomous cars goes and the closer we come to establishing the learning systems to retrace why a certain decision was made machine learning be secure?. Proceedings of the 2006 ACM Symposium on Informa- Lingjia; Mars, Jason. The Ar-chitectural Implications of Autonomous Driving: Cons- convenient technology into everday road traffic. („whiteboxing“ the system)? tion, Computer and Communications Security - (ASIACCS '06), March 2006. traints and Acceleration. Proceedings of the Twenty-Third International Conference on However, securing machine learning systems is not as easy as it may • Are comprehensible whitebox machine learning systems more vul- [2] Papernot Nicolas. Characterizing the Limits and Defen-ses of Machine Learning in Architectural Support for Programming Languages and Operating Systems (ASPLOS Adversarial Settings. May, 2018. '18), 751–766. 2018. seem. Most machine learning models are a blackbox, even to their nerable to attackers? developers. Decisions made by a model are often not comprehensi- Answers to such questions should raise OEMs‘, researchers‘ and ble by humans. developers‘ awareness to the security risks of machine learning sys- Thus, research in the area should give answers to the following ques- tems. Further research is encouraged to create machine learning tions: models more resilient against harmful attacks.

RESEARCH APPROACH AND METHODOLOGY

The first step in research is to determine which machine learning tions of a software architecture in autonomous vehicles [3]. algorithms will be used in autonomous vehicles. OEMs hardly Published attacks to machine learning models have to be evaluated give any information about their progress which presents a big and categorized to create a structured overview of the current state obstacle in any form of research in the field of autonomous cars of research. After that, the machine learning algorithms and any re- and machine learning. The issue is well known in the scientific spective attacks can be mapped to the machine learning algorithms community, which is why some researchers have made assump- used in autonomous vehicles. It is important to discuss transferable

82 | SYMPOSIUM SYMPOSIUM | 83 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

In the following, we will highlight some of the latest projects. They cover prototypes and applications that support efficient typing in VR The study reveals that with our apparatus, experienced typists ben- and an AR experience, that amplifies human,perception to foster a efited from seeing their hands, and reach almost outside-VR typing better understanding of a thermal flux experiment. performance. However, inexperienced typists profited most from semi-transparent hands, which enabled them to type just 5.6 WPM Typing in Virtual Reality: When we interact with a computing sys- slower than with a regular desktop setup. tem, entering text is one of the most common tasks. To enable users to work as efficiently in a virtual environment as in a real office they Learning in Augmented Reality: To support individual learning require high-performance generic input devices. Users that are im- skills and alleviate the workload of supervisors, modern MR systems Mixed Realities in Everyday Life mersed into a current VR environment usually do not perceive their have become an integral part of conveying knowledge within edu- environment including their own hands, nor the physical input de- cational institutes [4]. This includes the employment of MR systems, 08 vice. To investigate the different aspects of typing in a virtual environ- which support students just-in-time with additional individualized ment on a physical keyboard, we developed an apparatus that tracks content. With enhanced hardware and software solutions availa- the user’s hands, and a physical keyboard, and visualize them in re- ble, MR applications have finally been explored beyond the lab by Pascal Knierim al-time in the VR environment. Using our apparatus, the user can various researchers. We recently developed [10] and studied an AR perceive a virtual representation of their own hands and the physical experience that visualizes the current state of a physics experiment LMU Munich, Human-Centered Ubiquitous Media keyboard. Given these vis- and deployed it in a physics class. With our application, students get [email protected] ual cues, we enable the user to easily approach and type on the key- realtime feedback and enhanced data visualization of the experi- Prof. Dr. Albrecht Schmidt board. To evaluate the effects of avatar hand representation we ran ment. Fist results of a user study using the prototype showed a small a text input study with 32 participants. We investigated the overall positive effect of augmented reality on students’ performance with achievable text input speed, task load and felt presence. A detailed regards to acquired knowledge in thermodynamics [9]. description of the apparatus, as well as the study methodology and INTRODUCTION results, can be found in the publication [5].

Over the last decade, technology in the field of Virtual Reality (VR) and trolled environments. In future MR systems, we could efficiently Augmented Reality (AR) matured. Finally, handheld and head-mount- work and display information without being bound to rectangular ed displays to immerse oneself are available to the consumer mar- two-dimensional screens. Using MR technologies virtual and phys- FUTURE WORK AND OPEN CHALLENGES ket. Still, end-users mainly immerse themselves in VR and AR for lei- ical worlds would further blend together. The border between MR sure, consuming 3D movies and games provided by entertainment systems and the reality will probably vanish in everyday life scenar- Regarding the augmentation of the thermal flux experiment, we con- An interesting topic that could conclude the dissertation would be an applications. Advances to visual fidelity, environmental tracking and ios in the near future. To understand how we can use MR in these tinue our development to further optimize the user experience. Fur- exploration on how electronically recorded MR experiences needs the field of view (FOV) enable truly immersive experiences with great everyday life scenarios and how to create value we need to investi- ther, we plan to develop a handheld visualization of the very same to be presented to the user to relive them at a later point in time to potential. gate these new technologies, interaction concepts and meaningful experiment. Given that, we want to investigate how much of the real support active recall. Still, this is only a rough idea and needs further content creation. physical experiment including a probe, sensors, and the power sup- discussion. In the following, we will only refer to Mixed Reality (MR) to cover ply is necessary to actually impart the learning material. all-embracing AR or VR as proposed by Milgram et al. [7]. Most con- In this proposal, we will briefly describe two projects that contribute sumer MR systems and content are tailored for entertainment. How- to the overall frame of the planned dissertation. First, we present ever, in academia and industry, there is also a considerable amount a short summary of the findings on the effect of avatar hands on of serious experiences [2]. Still, almost none of these prototypes or typing performance in VR using a physical keyboard [5]. Further- REFERENCES MR experiences were studied or explored beyond the lab. more, we present a MR application amplifying the users’ visual sense to perceive thermal flux while conducting an experiment in physics [1] N. D. Finkelstein, W. K. Adams, C. J. Keller, P. B. Kohl, K. K. Perkins, N. S. Podolefsky, S. [7] P. Milgram, H. Takemura, A. Utsumi, and F. Kishino. Augmented reality: A class of Reid, and R. LeMaster. When learning about the real world is better done virtually: A displays on the reality-virtuality continuum. In Telemanipulator and telepresence tech- With the proliferation of MR HMDs to the general public and end-us- class [9]. study of substituting computer simulations for laboratory equipment. Phys. Rev. ST nologies, volume 2351, pages 282–293. International Society for Optics and Photonics, ers, it is nowadays feasible to study the potential of MR in less con- Phys. Educ. Res.,1:010103, Oct 2005. 1995. [2] M. Kesim and Y. Ozarslan. Augmented reality in education: current technologies and [8] M. E. C. Santos, A. Chen, T. Taketomi, G. Yamamoto, J. Miyazaki, and H. Kato. Aug- the potential for education. Procedia-Social and Behavioral Sciences, 47:297–302, 2012. mented reality learning experiences: Survey of prototype design and evaluation. IEEE [3] P. Knierim, F. Kiss, and A. Schmidt. Look inside: Understanding thermal flux through Transactions on Learning Technologies, 7(1):38–56, Jan 2014. augmented reality. In 2018 IEEE International Symposium on Mixed and Augmented [9] M. P. Strzys, S. Kapp, M. Thees, P. Klein, P. Lukowicz, P. Knierim, A. Schmidt, and PROBLEM STATEMENT AND RESEARCH QUESTION Reality (ISMAR-Adjunct), Munich, 2018. J. Kuhn. Physics holo.lab learning experience: using smartglasses for augmented [4] P. Knierim, T. Kosch, M. Hoppe, and A. Schmidt. Challenges and opportunities of reality labwork to foster the concepts of heat conduction. European Journal of Physics, mixed reality systems in education. In Mensch und Computer 2018 – Workshopband, 39(3):035703, 2018. Industry, academia and other stakeholders are constantly pushing A comprehensive analysis of MR learning applications has shown Dresden, 2018. [10] P. Knierim, F. Kiss, and A. Schmidt. Look inside: Understanding thermal flux through MR including hardware, interaction concepts, integration, and con- that these applications have diverse effects on the students’ perfor- [5] P. Knierim, V. Schwind, A. M. Feit, F. Nieuwenhuizen, and N. Henze. Physical keyboards augmented reality. In 2018 IEEE International Symposium on Mixed and Augmented tent creation. With the proposed research project, we try to inves- mance [8]. It is important to understand in which way we can inte- in virtual reality: Analysis of typing performance and effects of avatar hands. In Proc. of Reality (ISMAR-Adjunct), Munich, 2018. CHI ’18, pages 345:1–345:9, New York, NY, USA, 2018. ACM. tigate how the general public could benefit from ubiquitous MR in grate MR and how-to create MR experiences that blend into our lives [6] M. McGill, D. Boland, R. Murray-Smith, and S. Brewster. A dose of reality: Overcoming everyday life scenarios. The focus of research lies in particular in the to maximize the positive effects. With the planned research we seek usability challenges in vr head-mounted displays. In Proc. of CHI ’15, pages 2143–2152, New York, NY, USA, 2015. ACM. working and learn- to get a better un- ing domain. When MR technology becomes an integral part of our derstanding how to build meaningful learning and working environ- life we need to understand how to create relevant applications that ments and how to integrate MR in our lives. Additionally, we will elab- are intuitive to use. We try to investigate how MR will change our orate design guidelines that support the future development of MR work and education structure and which steps are necessary to inte- experiences. grate MR into these domains?

RESEARCH APPROACH AND METHODOLOGY

In the early stage of the projects, online surveys and semistructured studies in a controlled lab setting. However, we will also conduct stud- interviews will be applied to gather qualitative and quantitative data ies beyond the lab when the prototypes are mature fully developed. and ideas on how consumers would use MR. Later, prototypes will be To evaluate the effect of the built prototypes we will run preand post- built in an iterative process. These will be evaluated in user-centered test and analyze the data with the statistical tools needed.

84 | SYMPOSIUM SYMPOSIUM | 85 ABSTRACTS count. The second dimension is the presentation of the content. context for adaptation. For example, changing presentation based SYMPOSIUM The interface may adapt to the user's preferred language, their cur- on environmental conditions. The basic concept of adaptation to rent cognitive abilities by simplifying visualizations and choice of context is described in Schilit's seminal work [4]. We extend this words or environmental conditions by changing screen brightness. work by a more fine-grained, per-user customization that takes ad- Lastly, the interaction itself can be proficiency-aware by taking ditional dynamic properties of the user into account. into account the user's current abilities and environment to offer a suitable interaction method. In this last level, we have identified detecting a user's proficiency using ubiquitous sensing techniques. Here, the system employs Detecting Proficiency physiological sensing devices to infer the current state of the user, On an abstract level, this module specifies whether adaptation is such as cognitive ability and motoric skills. Example modalities Creating Proficiency-Aware Systems needed to cope with the user's lack or abundance of proficiency. include face detection, tracking of eye and hand movements. Sys- tems may enable proficiency detection down to individual users through Ubiquitous Sensing Technology Detecting a user's proficiency can be done on multiple levels. The depending on the scenario and employed modality, while still pro- most static one being user profiles. The system recognizes a spe- viding robust and coarse detection if necessary as we outline in the 09 cific user and adjusts based on their pre-defined settings. In public next section. spaces, a more realistic approach is to leverage the surrounding Jakob Karolus LMU Munich, Human-Centered Ubiquitous Media [email protected] · www.jakob-karolus.de PLANNED NEXT STEPS Prof. Dr. Albrecht Schmidt In this thesis, we present a definition of proficiency-aware systems trary to long-term user adaptation based on context-awareness, we and outline a generic architecture and design space for researchers strive to realize short-term proficiency detection by leveraging ubiq- to build upon. uitous sensing techniques. Within this area of our proposed design INTRODUCTION space, we continuously address the challenge to infer users' profi- Our own work entails the vision of ad-hoc adaptiveness when inter- ciencies and to provide appropriate adaptations. acting with ubiquitous systems throughout our everyday life. Con- The number of digital devices around us is steadily increasing. This is text-awareness are popular methods to adapt interactive systems. not only true for our own devices, but also for shared devices within With the proliferation of affordable sensing devices, researchers our environment. Examples include ATMs, ticket vending machines have started to leverage physiological sensing to infer user states and ordering kiosks at fast-food restaurants. Commonly, these sys- automatically, alleviating the need for explicit user input. PRELIMINARY RESULTS tems are designed to be easy to operate. However, the background and knowledge of people differs substantially, due to cultural as well We want to further elaborate on the idea of systems, that automat- In our work, we focus on ubiquitous sensing techniques to infer pro- Contrarily, in EMGuitar [2], we opted for a fine-grained proficiency as individual differences. Hence, interfaces that aim for a one-size- ically detect the users' proficiency. Therefore, we introduce the idea ficiency. So far, we addressed the user's language proficiency and detection by leveraging Electromyography to infer a user's skills fits-it-all approach will fail to address user-specific diversity which of proficiency-aware systems which describes systems that adapt to a leveraged robust gaze features to detect whether users were able to when playing the guitar. Here, we adapted the playback speed of our can impede operation. user's knowledge and skills. To this end, we formulate a definition for read the displayed language [3]. We showed that a binary detection, tutoring system, when the user is struggling to play correctly. The proficiency awareness as well as outline a generic architecture and a whether the user is able to read the language, can be done in under connection between low user proficiency and suitable adaptation is The solution to this problem has been around for decades: "Know suitable design space for proficiency-aware systems. The thesis work three second using one sentence only, allowing for quick adaptation. known beforehand as decreasing the tempo will allow better play for the user" [1]. The first principle in Hansen's list of user engineering focuses on populating specific aspects of this space. Exemplarily, we Yet, it is not possible to "guess" a more suitable language since this a novice. By automatically adapting playback speed during sessions principles for interactive systems is a simple idea, yet sometimes point out proficiency-aware systems from preliminary work. type of proficiency does not entail any ordering. Hence, other adap- based on previous play correctness, we realized the cycle between incredibly hard to achieve. In recent years, user profiles and con- tation methods, e.g. displaying a language selection menu, have to detection and adaptation for proficiency-aware systems as outlined be considered. before. PROBLEM STATEMENT AND RESEARCH QUESTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Everyday lives become increasingly reliant on digitally enhanced ar- privacy while providing a customized user experience. To establish tifacts. Users expect that devices are easy to use. For technologies terminology, we provide a definition of proficiency-aware systems, Ubiquitous sensing technologies are prone to environmental arti- venting the need for complex machine learning approaches. Here, deployed in public spaces and for shared devices adaptation to the elaborate on the idea by outlining a framework, describe require- facts and tend to be highly person-dependent. Hence, it is essen- we want to look into examples from other domains that require de- user is a key to achieve this. In this thesis, we introduce the concept ments, and point to examples from preliminary work. The research tial to identify levels of abstraction in the employed feature space, vice calibration for user interaction as an inspiration. of proficiency-aware systems, that can dynamically adapt to the users' challenge is in ubiquitous sensing techniques to infer users' profi- but also evaluate the possibility of short-term system calibration skills and expertise. Proficiency-aware systems have means to detect ciencies and to provide appropriate adaptations. We envision that schemes that seamlessly integrate into the user interaction, circum- the users' proficiency relevant to the task and can adapt the user for computing systems in public spaces this idea can lead to new interface and the content accordingly. Such adaptation can be done interaction paradigms. locally and does not require to know the user, hence preserving the REFERENCES

[1] W. J. Hansen. 1971. User Engineering Principles for Interactive Systems. AFIPS ’71 (Fall). [4] B. Schilit, N. Adams, and R. Want. 1994. Context-Aware Computing Applications. In RESEARCH APPROACH AND METHODOLOGY ACM, New York, NY, USA, 523–532. 1994 First Workshop on Mobile Computing Systems and Applications. 85–90. [2] J. Karolus, H. Schuff, T. Kosch, P. W. Wozniak, and A. Schmidt. 2018. EMGuitar: Assisting Guitar Playing with Electromyography. DIS ’18. ACM, New York, NY, USA, 651-655. Definition: Proficiency-Aware Systems are systems that can skills, background knowledge, motoric skills, and cognitive ability. [3] J. Karolus, P. W. Wozniak, L. L. Chuang, and A. Schmidt. 2017. Robust Gaze Features for Enabling Language Proficiency Awareness. CHI ’17. ACM, New York, NY, USA, sense a user's proficiency and adapt the content, the presentation, Every proficiency-aware system is governed by a continuous cycle 2998–3010. and the interaction. involving proficiency detection on the one side and appropriate adaptation on the other side. Both sides mutually influence each Hence, the term proficiency-aware system refers to a dynamic inter- other until an equilibrium is reached. connection between system and user, that continuously assesses the user's ability to interact with it and adapts accordingly. Such adap- Interface Adaptation tations will enable a user experience tailored to the specific users' We have identified three main dimensions of adaptation. Firstly, an proficiency. The proficiency hereby refers to the knowledge, skills, interface can adapt its content, in other words, it adapts towards and abilities that are relevant for using the system, such as reading the user's interests also taking the user's prior knowledge into ac-

86 | SYMPOSIUM SYMPOSIUM | 87 ABSTRACTS For creating true multilingual systems one can either use recent straightforward due to the lack of manually annotated data in many SYMPOSIUM attention mechanism such as self-attention [3] which are heavily languages. Thus, we fall back to unsupervised evaluation criteria based on word embeddings or deploy meta learning methods such and introduced for example roundtrip translation, a new evaluation as model agnostic meta learning [2]. We will first start with extend- method for multilingual word embeddings which does not require ing and adapting attention mechanism to our multilingual setting, as any manually annotated data. this method has already proved successful in the field of machine translation. Levy et al. [5] posed a great inspiration for our work, as they showed that for multilingual embeddings the actual embedding algorithm is Our research is mainly driven empirically. Implementing the pro- not as decisive as the features used as input to the algorithm. Thus, posed algorithms, deploying them on suitable datasets and evaluat- we introduced a new feature for embedding learning and proposed Effective Usage of Multilingual Data in ing the results rigorously is integral part of the project. However, as concept based embedding learning. 10 Natural Language Processing we work usually with a huge number of languages, evaluation is not PRELIMINARY RESULTS

A first conference paper [4] summarises our preliminary results. nisms called multi-attention. First result show slight improvements Philipp Dufter Here, we induced multilingual word embeddings based on the Bible on monolingual data. However, it would be easily scalable across lan- Ludwig-Maximilians-Universität München, Centrum für Informations- und Sprachverarbeitung – a resource which exhibits the greatest language coverage currently guages which is the main reason why we work on this new method. available. Now we are exploring an extension of attention mecha- [email protected] Prof. Dr. Hinrich Schütze

PLANNED NEXT STEPS INTRODUCTION In an immediate next step, we will focus on deploying the mul- outperforms mono- or bilingual approaches in applications such as There are roughly 7000 languages with around 700 being econom- addressed. For example, Wikipedia, one of the biggest natural lan- ti-attention mechanism to multiple languages. In the long term, machine translation or named entity recognition. Third, we need ically relevant (i.e., more than 1 million active speakers or official guage resources, covers only 291 languages. Thus, there is a huge in order to be able to summarise these findings in a dissertation to show that these methods are not only applicable for processing state language). While natural language processing (NLP) showed need and potential of scaling natural language processing methods there are several steps missing: first we need to integrate multi- across languages, but also across different domains. enormous progress over the past years and every day new impres- across languages. lingual word embeddings into specific tasks such as named entity sive applications are released, only a small number of languages is recognition. Second, we need to show that a multilingual approach

PROBLEM STATEMENT AND RESEARCH QUESTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

Currently, almost every NLP system (such as named entity recogni- legal texts is very different from the language used in tweets. Thus, There are a several open questions and we would appreciate dia- Ideas in identifying this kind of features are highly appreciated. tion, relation extraction or textual entailment) needs to be developed one can think of these two domains as two different languages (not logue and exchange in these areas: • Cross-domain learning is an active research area in many fields for every language separately. This is not only economically ineffi- from a linguistic, but from a processing point of view), which makes it • How does multilingual data in industry look like? We are in contact such as computer vision. We are highly interested in feedback on cient, but it does not allow for sharing information across languages obvious that cross-lingual methods are applicable for cross-domain with one industry partner but would be interested in getting addi- our methods or ideas for adapting existing methods for NLP. either. We intend to tackle this lack of transfer across languages by learning. The number of domains is far bigger than the number of tional data sources and/or opinions. researching multilingual approaches. languages and for industry application it is of utmost importance to • Language independent features (i.e. features which are the same have tools tailored to their domain in order to achieve acceptable across all languages, such as pictures, links, ids) are of great value. A solution to the above problem would be useful to both, academia performance. and industry. Establishing multilingual methods gives academia access to large amounts of data. Suddenly not only one Wikipedia is To give an example of a use case: ticketing systems for large cor- REFERENCES available, but 291 Wikipedias can be used simultaneously to improve porations aggregate a lot of information. Usually, this information is

NLP systems. Creating Machine Translation systems for low-re- spread across many languages and the language used in the tickets [1] Klementiev A. et al. "Inducing crosslingual distributed representations of words." [4] Dufter P. et al. "Embedding Learning Through Multilingual Concept Induction." source languages will be easier and typological analysis can benefit is a special domain relying on domain specific terminology. Multilin- Proceedings of COLING 2012. Proceedings of ACL 2018. from these insights, as well. Currently most NLP methods operate gual methods will facilitate the processing of this information and [2] Finn C. et al. "Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks." [5] Levy O. et al. "A Strong Baseline for Learning Cross-Lingual Word Embeddings from Proceedings of ICML 2017. Sentence Alignments." Proceedings of ACL 2017. on English only. This procedure exhibits great linguistic biases. Multi- can be for example used to build a system which retrieves similar [3] Vaswani A. et al. "Attention is all you need." Proceedings of NIPS 2017. lingual methods will help the research community to approach NLP tickets across all possible languages in order to suggest suitable solu- more objectively. tion approaches to an agent.

At the same time, industry will benefit greatly from multilingual The key research question to achieve the aforementioned objec- methods as players can scale across many markets more easily and tives is twofold: how can information be efficiently and effectively collaboration within globally operating firms is facilitated. In addition, exchanged across languages and how can this creating multilingual systems is closely related to creating systems information be used to increase the performance of NLP systems which perform well across different domains. The language used in across languages.

RESEARCH APPROACH AND METHODOLOGY

Regarding the information exchange across languages, there are ing knowledge across languages [1]. Our project started by inducing two possibilities: storing generic information or task-specific infor- multilingual word embeddings across more than 1000 languages [4]. mation. Both are materialized by storing (a subset of) parameters of We focused first on obtaining generic presentations. The underlying a machine learning model (e.g., an artificial neural network). In the reasoning is that task-specific information can be easily added by the field of NLP word embeddings are an established method of shar- NLP system which uses the embeddings to solve a specific task.

88 | SYMPOSIUM SYMPOSIUM | 89 SYMPOSIUM ABSTRACTS RELATED WORK

Ochiai et al. (2014) created visual graphics in mid-air by trapping levi- polation between adjacent levitation points. However, using this al- tating particles in a two dimensional grid of acoustic nodes. The grid gorithm the movement of the levitating particle is smooth only along was generated using standing waves, with signal emitted by four, the vertical direction, whereas on the horizontal planes the particle pairwise opposing transducer arrays. The drawback of using the moves in discrete steps of 0.5cm. standing waves trapping method to create a physical display would be that to ensure stable movement of particles along one dimension, Finally, in the paper by Marzo et al. (2015) the problem of finding a two opposing arrays are required. Hence to obtain smooth move- set of transducer phases which produces strong and accurate traps Dexterous Interaction with ment in 3D, six arrays would be needed, which would occlude the at the desired points of the acoustic field was solved with a BFGS op- visual field of the user. timizer. However, solving an optimization problem for each point is Levitating Interfaces computationally intense, e.g., for a medium two-array set-up it takes A method to move a levitating particle across an acoustic field bet- about 1s. In addition, they introduced the notion of holographic 11 ween two arrays, was introduced by Omirou et al. (2015). They de- acoustic elements, by which every levitation trap can be decompo- veloped the LeviPath algorithm which combines two basic patterns sed to a signature pattern and a focusing element. Thus, moving a Viktorija Paneva of movement: a height variation and a 2D path obtained by inter- trap is equivalent to refocusing a holographic acoustic lens. Poster Faculty of Mathematics, Physics and Computer Science, University of Bayreuth, Serious Games Award PRELIMINARY RESULTS [email protected] · www.ai8.uni-bayreuth.de 1. PREIS Prof. Dr. Jörg Müller 1500€ So far, a setup of two transducer arrays with built-in cooling system The system was tested in the first ever Fitts’ Law levitation study, and adjustable rails has been built, surrounded by a motion cap- where users moved a levitating particle (cursor) between two need- ture system, able to track the movement of the levitating particle. le pointers (targets) in the acoustic volume (display). The study re- Next a modified version of the algorithm in Marzo et al. (2015) has vealed positive results, namely, users reached performance similar INTRODUCTION been implemented, which stores the phase values in a look-up ta- to that of a mouse cursor. The results of the study are important ble. The advantage is that the optimal phases for each point in the because they open the possibility for using a levitated particle as The project is about utilizing the potential of ultrasound technology analogous to pixels on a display, is suspended in air using ultrasonic field can be called within few microseconds, which makes real-time an improved cursor for mixed reality scenes, due to its ability to in the field of Human-Computer Interaction (HCI), for developing a waves, emitted by two arrays of phased transducers. The advantag- interaction with the display possible. provide depth cues. novel physical interface. This technology shall encourage the user es of the levitating display include: direct manipulation of physical to reach into a new type of display composed of levitating matter; visualizations, interaction with the bare hands (i.e. without devices and see, hear, feel and manipulate 3-dimensional objects in mid- likejoystick, gloves, gamepad etc.) and transparency, which enables air. The levitating matter composed of individual levitating particles, good visibility from many angles and encourages collaboration. PLANNED NEXT STEPS

PROBLEM STATEMENT AND RESEARCH QUESTION The project is in the beginning stage and there are still many avenues zo et al., 2015). Next, the usability of a levitating cursor in augmented to be explored. In the forthcoming stages, I would like to move away reality (AR) and virtual reality (VR) environments needs to be tested. from using a lookup table, due to the extensive computational effort Finally, multiple modalities are to be integrated in the new iterations The use of ultrasonic technology in HCI is novel and largely unex- controlling the acoustic field in a manner that enables fast, smooth required and aim for achieving the same or better performance with of the interface and the developed interaction techniques need to be plored. This research will provide valuable output for assessing its and stable movement of the levitating particles. To this end, it should a simpler algorithm, e.g., using holographic acoustic elements (Mar- validated in further user studies. potential, as well as its limitations. The prospects of fulfilling one of be determined what parameters of the system influence the stabi- the longest standing visions in HCI of the ‘Ultimate Display’ (Suther- lity of the particle movement the most. In addition, it is important land, 1965) are very exciting. Sutherland, envisioned a room where to identify and test which interaction techniques are best suitable a computer can control the existence of physical matter e.g. create for levitating displays. Finally, in the later development stages, the OPEN ISSUES AND POSSIBILITIES FOR COOPERATION and dissolve chairs etc. The Ultimate Display is one that would truly challenge would be to incorporate feedback, produced by multiple merge the physical and virtual world, introducing a radical shift in modalities, such as haptics, directional audio and visual projections. In the initial stages of the project, priority was given to building the In addition to young children, the benefit of a physical levitating in- the way people interact with computers. In this scenario, interaction There are many potential benefits of improving the currently avai- hardware and solving technical issues. In the future, I would like to terface can be explored among the elderly or people with disability. with computers will come very near to the intuitive and natural inter- lable prototypes and further developing the levitation system, e.g., prototype custom levitation interfaces for different use cases. Furthermore, the novelty factor of it makes it attractive for public action with the physical world. improved sense of agency, better overall user experience, more exhibitions (for example in a gallery or a museum). versatility in the manner the data is displayed and in the interaction One application area that can benefit from a levitating 3D interface is The main challenge in the development of the levitating display is techniques, and broader field of potential applications. for example early childhood education – in the study of Yannier et al. Interesting collaboration topics can arise with research groups wor- (2016) it was tested among 4-8-year-old children whether adding 3D king with mixed reality scenes, estimating and analyzing cognitive physical objects to mixed reality scenes provides better learnability workloads as well as using computational methods to improve well- RESEARCH APPROACH AND METHODOLOGY and enjoyment compared to 2D flat screen interaction. The results being. showed that observing physical phenomena in mixed-reality games leads to better learnability and enjoyment compared to a screen only The research methodology is composed of hardware design, inter- from the research on possible interaction techniques, in turn, pro- version. action techniques, demonstrators and user studies. The methods vide guidance for the subsequent iterations of the technology. Thus are carefully selected because they intertwine and complement each the mixed methodology approach hedges the technological and other. The hardware design enables the development of different scientific risk associated with using an “unconventional” technology interaction techniques and testing of those with user studies. The for interface design and grants agility and prompt realignment. Mo- REFERENCES user studies on feedback accuracy provide valuable information on dels and prototypes are frequently built with the purpose to explore hardware requirements as well as system constraints. The results the complex design space. [1] Marzo, A., Seah, A. S., Drinkwater, W. B., Deepak, R. S., Long, B., & Subramanian, S. [4] Sutherland, I. (1965). The Ultimate Display. Multimedia: From Wagner to Virtual Reality, (2015). Holographic Acoustic Elements for Manipulation of Levitated Objects. Nature 506-508. Communications. [5] Yannier, N., Hudson, S., Wiese, E., & Koedinger, K. (2016). Adding Physical Objects to an [2] Ochiai, Y., Hoshi, T., & Rekimoto, J. (2014). Pixie dust: graphics generated by levitated Interactive Game Improves Learning and Enjoyment: Evidence from EarthShake. ACM and animated objects in computational acoustic-potential field. ACM Transactions on Trans. Comput.-Hum. Interaction. Graphics. [3] Omirou, T., Marzo, A., Seah, S. A., & Subramanian, S. (2015). LeviPath: Modular Acous- tic Levitation for 3D Path Visualisations. CHI '15 Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems, 309-312.

90 | SYMPOSIUM SYMPOSIUM | 91 SYMPOSIUM ABSTRACTS RELATED WORK

In [3], 3D poses are used for intention and pose prediction of pe- before entering the street. However, this approach does not fore- destrians. For this, Quintero et al. use Balanced Gaussian Process cast the trajectory, the used dataset is small and contains only pe- Dynamical Models and a naive-Bayes classifier. However, a quanti- destrians. tative evaluation is performed solely on a high frequency and low noise dataset, which was recorded indoors. Fang et al. [4] use 2D A related area of intention detection of VRUs is the field of human poses of pedestrians estimated by [5] for detection of movement activity recognition without regard to application in the transport primitives from a moving vehicle. A Random Forest and a Support sector. For example, Kim et al. [6] use 3D human poses as input of Vehicle-Based Intention Detection Vector Machine based on features obtained from 9 joint positions a temporal CNN to recognize 60 different activities such as sitting describing the pedestrian's torso and legs are used. They are able down or hand waving. 12 of Vulnerable Road Users to anticipate the intention of a pedestrian to cross the road 0.75 s PRELIMINARY RESULTS Viktor Kress With the help of a test vehicle equipped with appropriate sensors mation, we found larger errors for cyclists compared to pedestrians University of Applied Sciences Aschaffenburg, Faculty of Engineering, Cooperative, Automated Traffic Systems and state-of-the-art methods for 2D and 3D pose estimation [5, 7], and decreasing performances with increasing distance. However, [email protected] trajectories of pedestrians and cyclists including their 2D and 3D the results do not deteriorate rapidly, showing that pose estimation poses could be extracted from real scenes in road traffic. To verify is applicable in transport sector. Prof. Dr.-Ing. Konrad Doll · Prof. Dr. Bernhard Sick the quality of these poses, a dataset with manually labeled poses was created. The evaluation showed that the quality of the 2D It is expected to achieve an exact and reliable intention detection by poses is high despite difficult conditions such as large distances, considering the poses. In addition, the identification of characteris- occlusions, varying light and weather circumstances. Furthermore, tic movement sequences of individual body parts and early indica- INTRODUCTION the obtained errors concerning 3D poses are sufficiently small in tors for movement primitives is intended. Understanding human relation to body size to characterize respective body poses and to poses and body language can even lead to a new kind of interaction In today’s worldwide traffic, 49% of the fatalities hit pedestrians, as emergency braking. allow the use for intention detection. For both, 2D and 3D pose esti- between humans and automated vehicles. cyclists, and motorcyclists [1]. Protection of vulnerable road users Human body pose contains relevant indications of the intention (VRUs), e.g., pedestrians and cyclists, needs to be improved by Ad- and future trajectories of VRUs. For example, [2] shows that start- vanced Driver Assistance Systems or automated driving functions. ing motions of cyclists are often preceded by arm movements. Due During situation analysis, intention detection of VRUs is necessary to the recent progress in human pose estimation, the use of body to identify potentially dangerous situations. The earlier they are poses for intention detection of VRUs in road traffic is conceivable. PLANNED NEXT STEPS recognized, the more time is available for countermeasures such After the ability to gather trajectories including 2D and 3D poses of this data. This includes the classification of movement primitives as VRUs in real traffic from a moving vehicle has been achieved, the well as the forecast of trajectories. This should provide insights and PROBLEM STATEMENT AND RESEARCH QUESTION next step is to develop models for intention detection based on requirements for future intelligent data collections.

To increase the safety of VRUs in road traffic, fast and reliable in- entire body pose should be considered. Therefore, the body pose tention detection is essential for use in Advanced Driver Assistance of VRUs needs to be estimated in real road traffic with sufficient ac- OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Systems and for automated driving functions since vehicles in ur- curacy using suitable sensors. The poses can be estimated two-di- ban traffic share the traffic routes with VRUs. Intention detection mensionally, i.e. in the form of joint positions in images, or three A major issue is the recording of balanced datasets, which can serve Cooperations may be helpful in the areas of environment percep- contains two objectives: Besides the early classification of move- dimensionally, i.e. in the form of real metric positions of key points. as a basis for machine learning methods. Interesting data, such as tion and object tracking as this is a requirement for the elicitation of ment primitives, such as starting of a pedestrian or turning of a Large distances, occlusions caused by objects or by the body of the dangerous situations for VRUs or unusual behavior, are rare but of the necessary data. Also, an exchange regarding Vehicle-2-X com- cyclist, it includes the trajectory forecast for the next few seconds, respective VRU itself and varying environmental conditions make special interest for the described application. munication would be interesting and an application of the achieved taking previous movements of the VRU into account. In collision this a challenging task. results in the area of path planning would be desirable. scenarios between vehicles and VRUs, intention detection enables countermeasures such as emergency braking or evasive maneu- From the perspective of a vehicle, VRUs are often only visible for a vers to be initiated earlier than the human reaction time and atten- short time before a reaction of the vehicle may be necessary, es- REFERENCES tion would allow. Furthermore, path planning of automated vehi- pecially in critical situations. This raises two questions: First, how cles depends on the trajectory forecast of VRUs. can the trajectory of a VRU be reliably forecasted despite the short [1] World Health Organization, “Global Status Report on Road Safety 2015,” 2015. [5] Z. Cao, T. Simon, S.-E. Wei, and Y. Sheikh, “Realtime multi-person 2d pose estimation observation time and second, how can the perception be extended [Online]. Available: http://www.who.int/violence_injury_prevention/road_safety_sta- using part affinity fields,” in IEEE Conference on Computer Vision and Pattern Recogni- Representative data are required to model the behavior of VRUs. beyond a single vehicle. For this purpose, appropriate methods for tus/2015/en/ [Accessed: 04/30/2018] tion (CVPR), 2017, pp. 1302–1310. [2] A. Hubert, S. Zernetsch, K. Doll, and B. Sick, “Cyclists’ starting behavior at intersec- [6] T. S. Kim and A. Reiter, “Interpretable 3d human action analysis with temporal convo- This raises the question of how these can be intelligently collected data fusion between vehicles or vehicles and infrastructure, such tions,” in IEEE Intelligent Vehicles Symposium (IV), 2017, pp. 1071–1077. lutional networks,” in IEEE Conference on Computer Vision and Pattern Recognition so that, for example, critical situations and different VRU types re- as intersections equipped with sensors, must be developed. This [3] R. Quintero, I. Parra, D. F. Llorca, and M. A. Sotelo, “Pedestrian intention and pose pre- Workshops (CVPRW), 2017, pp. 1623–1631. ceive sufficient coverage while the necessary effort for recording cooperative intention detection of VRUs shall improve the safety diction through dynamical models and behaviour classification,” in IEEE International [7] D. Tome, C. Russell, and L. Agapito, “Lifting from the deep: Convolutional 3d pose Conference on Intelligent Transportation Systems (ITSC), 2015, pp. 83–88. estimation from a single image,” in IEEE Conference on Computer Vision and Pattern remains as low as possible. especially in places with high traffic volumes. [4] Z. Fang, D. Vázquez, and A. M. López, “On-board detection of pedestrian intentions,” Recognition (CVPR), 2017, pp. 5689–5698. To be able to detect the intentions of VRUs early and reliably, the Sensors, vol. 17, no. 10, 2017.

RESEARCH APPROACH AND METHODOLOGY

This research uses a quantitative, data-driven approach. It tries validated and compared to other research using public datasets. to gain new insights from the collected data. Therefore, statistical If no suitable data exist, an appropriate dataset should be created methods and machine learning techniques are used. All findings and published. In particular, application orientation will be ensured are evaluated based on real data and should be application-orient- through cooperations with industry. ed. In order to ensure the quality of this work, the results will be

92 | SYMPOSIUM SYMPOSIUM | 93 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

The research approach is structured as depicted in Figure 1. In With respect to content design, and thus research question (B), the analysis phase, currently available tools and research are ex- Walker’s article on Memrise points out several key aspects such amined. Moreover, this is where knowledge from different areas as using a “variety of methods” and to integrate tests [6]. Howe- is combined in order to highlight opportunities for improvement. ver, she only investigates vocabulary learning. Cognitive state and the memory of learners, learning strategies and success, but also the context and learning environment are In the development and evaluation phase, on the other hand, the important aspects to be considered here. A specific task will be insights from the analysis phase will be used to develop impro- It’s All Greek to Me – Support Language to explore the potential of ubiquitous technologies such as Aug- ved learning technologies and to evaluate them in user studies. mented Reality (AR) and wearable devices (e.g. smartwatches), Finally, it has to be noted that this two-phase approach is not Learning with Ubiquitous Technologies as these are very relevant for research question (A). Notable re- meant to be strictly linear. Rather, it is an iterative process, whe- lated research in this area is Dingler et al.’s work on microlear- re findings from the experimental part are then used as basis for 13 ning sessions in a mobile context [2], and Goodwin-Jones’s work further analysis and subsequent developments. on AR for vocabulary learning [3]. Fiona Draxler Ludwig-Maximilians-Universität München, Human-Centered Ubiquitous Media PRELIMINARY RESULTS [email protected] · www.um.ifi.lmu.de The research of this thesis is still at a very early stage. However, I late my research questions. Furthermore, I have looked at various Prof. Dr. Albrecht Schmidt have already gathered and analyzed a large number of publications language learning apps for different languages and tested state-of- from different areas which, amongst others, allowed me to identify the-art AR and Virtual Reality devices to get a sense of how they can the five issues described in the problem statement and to formu- be used. This experience will be a valuable guidance in the future. INTRODUCTION

While English has been established as a de-facto standard for com- cation elements aim to make them entertaining and motivate users munication in many areas, there are still many situations where to stay on track. Moreover, basic versions are often free of charge PLANNED NEXT STEPS knowledge of a local language is very helpful – just imagine being and sessions can range from a couple of minutes to several hours, hospitalized in a foreign country. Furthermore, learning other lan- as fits the user. As mentioned above, I am currently working on structuring the in- Another point on the agenda is the development of a first proto- guages is essential for a more profound understanding of a foreign formation collected so far. The identified trends will then be quan- type, where the general idea is the augmentation of physical books culture. In general, the emergence of pervasive and other new technologies tified and summarized in the form of a systematic literature review. to, for instance, provide explanations for unknown words in such a opens up opportunities for language learning, but there are also way that disruptions caused by the interaction are minimal. Nowadays, many language learners use apps and other technologies challenges to be mastered in order for them to contribute to suc- for individual study instead of traditional methods such as classroom cessful learning. In my work, I look at which ubiquitous technologies instruction. Language learning apps such as Duolingo and Memrise it makes sense to integrate and how this should be done to support OPEN ISSUES AND POSSIBILITIES FOR COOPERATION advertise an easy way of quickly developing language skills. Gamifi- the learning process.

The field of language learning can be explored in many different di- from others could be helpful for identifying the most promising directions and even after defining the two research questions above, rections for further research and projects. PROBLEM STATEMENT AND RESEARCH QUESTION there are still many options for a specific path to follow. Thus, input

To a certain extent, language learning apps are very useful. However, there are also several shortcomings that need to be considered: I will, in particular, consider aspects (4) and (5) and address the fol- REFERENCES 1. The apps usually focus on specific aspects such as vocabulary, and lowing questions: then neglect grammar instruction. (A) How can technology support language learners where and when 1] Coniam, David. “The linguistic accuracy of chatbots: Usability from an ESL perspective”. [4] Heil, Catherine et al. “A Review of Mobile Language Learning Applications: Trends, 2. They cannot replace conversation with actual people. it is most appropriate for them? Text and Talk, 34(5) 545-567. 2014. Challenges, and Opportunities”. The EuroCALL Review, 24(2) 32-50. 2016. 3. They can only provide a superficial glimpse into the foreign cul- (B) How can the content presentation and assessment be designed [2] Dingler, Tilman, et al. “Language Learning On-The-Go: Opportune Moments and [5] Seedhouse, Paul, et al. “The European Digital Kitchen Project”. Bellaterra Journal of Design of Mobile Microlearning Sessions”. Proceedings of the 19th International Teaching and Learning Language and Literature, 7(1) 1-16. 2014. ture, as there is no physical or immersive experience. to cater to cognitive demands in general, but also to individual Conference on Human-Computer Interaction with Mobile Devices and Services. ACM, [6] Walker, Louise. “The impact of using Memrise on student perceptions of learning Latin 4. They are not sufficiently personalized to adapt well to the user’s needs? 2017. vocabulary and on long-term memory of words”. Journal of Classics Teaching, 16(32) individual skills and knowledge, as well as to their usage habits [4]. [3] Goodwin-Jones, Robert. Augmented Realities and Language Learning: From Annotated 14-20. 2016 Vocabulary to Place-Based Mobile Games. Language Learning & Technology, 20(3) This is particularly critical for learners who apply several learning 9-19. 2016 methods that are independent from one another. 5. The way quizzes and activities are designed, they often promote passive recognition rather than active recall. Current technologies provide the means to tackle some of these issues. For example, chatbots are already good enough to mimic simp- Figure 1 – Phases of the Research Approach le conversations [1] and machine learning approaches are likely to lead to further improvements in the future. In addition, there are projects aiming to create a cultural experience around the language learning part, for instance in the European Digital Kitchen, where learners cook a typical dish in an augmented environment [5].

94 | SYMPOSIUM SYMPOSIUM | 95 ABSTRACTS we build on and that to some extend we would like to communicate portunities for questioning established methods and believes that SYMPOSIUM to users of biometric systems; on the other hand they also open op- have been accepted as general truth in the field for years.

PRELIMINARY RESULTS

As a first step towards informing users about the strength of be- tems understandable is an important step to foster the adoption of havioural biometric techniques we presented a design space for biometric systems in its early stages. Form our still running 4 week security indicators for behavioural biometrics revolving around the field study we expect insights towards user willingness to adopt Informing users’ decision making towards dimensions, input, output and purpose at CHI 2018 [7]. In this work continuous authentication systems and the impact of our new de- we also proposed example designs for such indicators, e.g. based sign on annoyance caused by the need for re-authentication. the adoption of biometric techniques on the features currently used or supporting the decision for specific biometric techniques during enrolment. One further field of current exploration is the extent to which users 14 are able to control their behaviour. This is in particular important, Further results are to be expected on the field of transparency of as the use of behavioural biometrics in general is built on the prem- continuous biometric systems. Such systems are based on an in- ise that 1) behaviour is unique to a user and 2) hard to imitate. Lukas Mecke ternal score, indicating that the current user is indeed the owner With this work we aim to accept or reject this premise for the case of the respective device. If this score is insufficient, users need to of typing behaviour on a smartphone. Furthermore, we explore if University of Applied Sciences Munich, Biometrics++ verify their identity by re-authenticating. To this field we contribute there are differences between users with respect to their ability to [email protected] a novel design to 1) inform the user about the current confidence control their behaviour which would hint both at potential differ- Prof. Dr. Florian Alt score of the system and 2) provide announcement for the need for ences of security for those users as well as the feasibility of new re-authentication. While we hope this will be less of an issue for authentication mechanisms where users actively control their be- future systems (due to increasing accuracy), making biometric sys- haviour for authentication. INTRODUCTION

The use of biometric techniques holds a lot of promises for user au- unique physiological or behavioural traits). The aim of this work is to PLANNED NEXT STEPS thentication but also for applications beyond authentication. Howev- inform user decision making in the context of biometric techniques er, this is a rather novel field – at least from a consumer perspective based on personal features; that is to evaluate how user specific This work is conducted under the premise that users have miscon- obfuscated data) in order to determine how unique that trait is and – and we hypothesize there are several misconceptions amongst us- traits influence the strength of biometric systems. Furthermore we ceptions about biometric systems and their functionality. As a next consequently, how likely it is that someone might gain access to the ers towards the functionality and practical security of biometric and aim to foster transparent user information on this topic. step we plan to verify this hypothesis. This also serves the purpose protected device by chance. As we believe there is also an impact in particular behavioural biometric systems (i.e. systems that use of exploring how those misconceptions actually look like in order to of the users’ current context (e.g. a face might be very unique when be able to address them better. Another planned project aims to in- compared to other individuals in a small town whereas this is not dividualize feedback on the security of biometric techniques as men- the case for a whole country or some other town) we aim to take this PROBLEM STATEMENT AND RESEARCH QUESTION tioned before. In particular we plan to compare biometric traits of a factor into consideration as well. single user to the whole population (either based on generated or In contrast to traditional authentication methods like passwords or ty of (behavioural) biometric systems, not only as previously done for tokens, the strength and success of biometric and in particular behav- a whole population (i.e. by giving error rates) but also for individual ioural biometric techniques is not objectively measurable but highly users. Tightly correlated to this question are further investigations to- OPEN ISSUES AND POSSIBILITIES FOR COOPERATION dependent on the user. To judge the strength of passwords there are wards the influencing factors on individual security and explorations approaches like password meters (e.g. [1, 2]). No such thing exists for towards how to best communicate security and risks of the adoption This work is built on several premises, namely 1) users have miscon- approach we proposed above might only be one of them. Thus any biometric techniques, i.e. there is no mechanism in place to inform of biometric techniques (i.e. consequences if the biometric trait is lost ceptions about biometric systems and 2) there is indeed a way to input in that direction would be highly appreciated. This also opens users if their biometric traits (i.e. their particular physiology or behav- or stolen/faked) to a potential user. quantify security of biometric traits for individuals. Especially for the the opportunity for cooperation with experts from the related fields iour) are well suited to be used for biometric identification. Howev- last point there might be multiple ways to achieve this goal and the of cryptography, machine learning and psychology. er, as biometric traits cannot be changed, this information would be Thinking more general, advances in this field would enable a new view much needed to inform user decision making towards the adoption of on biometric systems that is less holistic and more individual which biometric techniques. should in the long run benefit the users of such systems. Also, a deeper understanding of what user traits influence the strength of biome- REFERENCES Based on that, our main research question is how to quantify securi- tric systems has the potential to inform the design of future systems. [1] Komanduri, S., Shay, R., Cranor, L. F., Herley, C., & Schechter, S. E. (2014, August). [5] Yampolskiy, R. V., & Govindaraju, V. (2008). Behavioural biometrics: a survey and Telepathwords: Preventing Weak Passwords by Reading Users' Minds. In USENIX classification. International Journal of Biometrics, 1(1), 81-113. Security Symposium (pp. 591-606). [6] Jain, A. K., Ross, A., & Prabhakar, S. (2004). An introduction to biometric recognition. [2] Ur, B., Alfieri, F., Aung, M., Bauer, L., Christin, N., Colnago, J., ... & Johnson, N. (2017, IEEE Transactions on circuits and systems for video technology, 14(1), 4-20. RESEARCH APPROACH AND METHODOLOGY May). Design and evaluation of a data-driven password meter. In Proceedings of the [7] Mecke, L., Prange, S., Buschek, D., & Alt, F. (2018, April). A Design Space for Security 2017 CHI Conference on Human Factors in Computing Systems (pp. 3775-3786). ACM Indicators for Behavioural Biometrics on Mobile Touchscreen Devices. In Extended Ab- [3] Ballard, L., Lopresti, D., & Monrose, F. (2007). Forgery quality and its implications for stracts of the 2018 CHI Conference on Human Factors in Computing Systems (p. LBW003). Within this project we aim to use a wide range of research methods, of which we would like to point out the work by Ballard et al. [3] behavioral biometric security. IEEE Transactions on Systems, Man, and Cybernetics, Part B ACM. including but not limited to surveys (to assess general attitudes of in particular. In their work they not only showed that signatures (Cybernetics), 37(5), 1107-1118. [4] O'Gorman, L. (2003). Comparing passwords, tokens, and biometrics for user authenti- users), lab studies (to generate reliable data) and field studies (to are prone to forgery but also that users had severe misconceptions cation. Proceedings of the IEEE, 91(12), 2021-2040. observe real world impact of our ideas). While it might be possible about how easy it would be. We believe this work motivates very to tackle some of our questions on a theoretical level, we believe well, that - beyond finding possible security risks – informing us- that as soon as actual users are involved in using a system (espe- ers should be a mayor goal. In this project we aim to take a step cially one that is based on user traits as is the case for biometric back from targeted attacks on biometric systems and explore mis- systems) it is vital to take human factors into consideration and conceptions about general security of biometrics (e.g. the risk of a test any proposed solution with real users. This is also the general stranger gaining access by chance) and possible countermeasures. standard in the field of Human-Computer-Interaction (HCI) in which Furthermore, we take inspiration from classical works in the field we operate with this project. of biometrics [4-6],

We find motivation for our work at various locations. For one, there describing biometric properties, their use and their classification. are a multitude of papers describing attacks on biometric systems These works are one the one hand the theoretical foundation that

96 | SYMPOSIUM SYMPOSIUM | 97 ABSTRACTS ed and less-affected body sides, particularly between free-living SYMPOSIUM and therapy. During free-living, postures below the shoulder were often estimated in both body sides. In contrast, during clinician guided therapy, postures including above shoulder movements Fig. 2: Functional range of were reached frequently, indicating patients' movement potential motion analysis of upper arms relative to the shoulder joint [3]. using orientation estimation and posture cubics. Wearable sensor-based motion analysis and recovery trend quantification of hemiparetic patients in free-living DISCUSSION 15 Wearable computing could enable digital healthcare. Our research, urement periods likely, hence predicting further recovery progress including wearable sensors and subsequent motion data analysis might be feasible. Walking segment extraction and walking parame- derived from hemiparetic patients revealed valuable insights in the ter analysis could lead to better understanding of patients' mobility Adrian Derungs stroke rehabilitation process and how a recovery trend could be behaviour. Moreover, evaluating physical activity, to promote an ac- quantified. The observation study in the day-care centre enabled tive lifestyle by providing exercise recommendations and quantified Friedrich-Alexander University Erlangen-Nürnberg, Faculty of Medicine data analysis and investigation of different approaches, useful to feedback is an option [2]. Thus, health and well-being might improve. [email protected] · www.cdh.med.fau.de devise further research. In particular, the observation study, where Beside patients and clinicians, health insurers could benefit from the Prof. Dr. Oliver Amft no interventions were intended, and patients could follow their in- patient digitalisation and the continuous behaviour information ex- dividual therapy programmes, real-life conditions patients will likely traction based on wearable computing. face after clinical rehabilitation were created. Hence, we believe our Although wearable sensors and computing render the digitalisation methods could be applied in free-living. For example, unsupervised of healthcare applications feasible, how such applications general- activity primitive extraction could be used to continuously assess ise among hemiparetic patients or across different cognitive impair- INTRODUCTION clinical scores daily, e.g. at home [1]. Such continuous quantification ments, e.g. Parkinson's disease, requires further research. renders personalised recovery trend evaluation over long meas- Wearable computing has become an ubiquitous technology in recent even years, thus continuous, objective measurement methods and years, thus enabling digitalisation of personalised healthcare to im- evaluation strategies are needed. We focused on longitudinal meas- prove health and well-being. In particular, digitalisation is beneficial urements and monitoring of patients during activities of daily living for patients, clinicians and insurers to derive clinical relevant infor- using IMUs to investigate patients' recovery after stroke without spe- mation, devise therapy strategies and provide health recommenda- cific test. In particular, comparing affected and less-affected body PLANNED NEXT STEPS tions. Moreover, patient-monitoring and recovery trend estimation sides could reveal insights, enabling clinicians to devise targeted in free-living becomes feasible. However, objective quantification therapy recommendations and intervention planning. Our research We analysed stroke patients' movement data using wearable IMUs measurements, are currently evaluated. Simulations offer potential and evaluation of patient-behaviour is challenging, especially for and methods could influence future stroke rehabilitation research, over several months and investigated recovery during free-living and for novel analysis approaches and evaluation of machine learning hemiparetic patients, e.g. after stroke, suffering individual functional rendering large scales studies in free-living feasible. In particular, we therapies. We showed insights in patient behaviour, and confirmed algorithms without the need of laborious data recording. Function- limitations. provide the following contributions: published related work. However, recovery trend modelling and al limitations could be investigated, and no burden for patients and 1. We show how clinical scores can be estimated without specific outcome predictions remain challenging. New simulation method- clinicians exist. In this research we used wearable sensors, i.e. inertial measure- tests using unsupervised methods, thus demonstrate how our ologies and synthetic sensor data analysis to complement real-life ment units (IMU) to monitor and describe patient behaviour, esti- research could be applied in remote monitoring applications in mate clinical scores, compare body sides, and quantify recovery free-living. trends without specific tests or restricting patients' daily life. So far, 2. We analyse upper and lower body parts of affected and less-affect- existing approaches as presented, e.g. by Patel et al, van Meulen ed body sides, to gain insights in patients movement behaviour. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION et al., to evaluate stroke patients' functional ability were limited to 3. We demonstrate how ubiquitous wearable sensors could be used short measurement durations ranging from minutes to hours, and for stroke recovery evaluation and proposed convergence points We are interested in cooperations, including clinical partners, and implementing follow-up studies are of interest to further evaluate required patients to perform specific functional tasks. Stroke reha- and posture cubics as quantification metric. researchers working in healthcare and rehabilitation related to neu- recovery trends in the home-environment of patients after stroke. bilitation, however, is a gradual process over weeks, months, and rological diseases, e.g. stroke and Parkinson's disease. The ZD.B-net- Currently, we explore additive manufacturing, 3D-printing technolo- work offers great potential of finding partners in research and econ- gies, and sensor-based human-computer-interaction using assistive omy. Cooperations could include the development and evaluation robotics. Our projects offer interesting challenges to research the of wearable computing systems integrated in ubiquitous accessory healthcare digitalisation, hence interesting for partners across the PRELIMINARY RESULTS or clothes. Approaches for personalised healthcare recommenda- ZD.B-network. tions and feedback system are intended. In addition, designing and We designed and implemented an empirical observation study We demonstrated, so far, that wearable sensors are suitable to an- during November 2013 and May 2014 in the neurological day-care alyse motion data of hemiparetic patients in free-living. We showed centre of the rehabilitation clinic Rheinfelden, Switzerland. We re- that the Extended Barthel Index, reflecting patients ability for inde- corded 102 full-day motion data, using six body-worn IMUs from pendent living, could be estimated with an average relative error of REFERENCES eleven hemiparetic patients during their individual rehabilitation 12.5%, compared to the clinical assessment, using support vector [1] Derungs, A., Seiter, J., Schuster-Amft, C., Amft, O., Estimating physical ability of stroke [3] Derungs, A., Seiter, J., Schuster-Amft, C., Amft, O., A metric for upper extremity func- programme. We evaluated approaches for unsupervised activ- regression [1]. patients without specific tests, Proceedings of the 2015 ACM International Symposium tional range of motion analysis in long-term stroke recovery using wearable motion ity primitive extraction from motion data and estimated clinical on Wearable Computers, ISWC, pp. 137-140 sensors and posture cubics, Proceedings of the 2018 IEEE International Conference on scores using machine learning [1]. We further showed that walking parameters derived from thigh- [2] Derungs, A., Schuster-Amft, C., Amft, O. Longitudinal Walking Analysis in Hemiparetic Wearable and Implantable Body Sensor Networks, BSN, pp. 17-20 Patients Using Wearable Motion Sensors: Is There Convergence Between Body Sides?, worn IMUs were suitable for walking segment detection across pa- Frontiers in bioengineering and biotechnology, vol.6, 2018 tients with high inter-patient walking differences. Moreover, we derived average specificity above 94%, demonstrating that non-walking activities were excluded. However, our convergence point analysis revealed that patient-individual habits and therapy schedules, influenced walking behaviour, rendering trend evaluation challenging [2].

The functional range of motion analysis using our intuitive 3D pos-

Fig. 1: Longitudinal convergence point estimation for recovery trend ture cubic metric, showed that upper arm postures differ in affect- evaluation in hemiparetic patients using walking parameters.

98 | SYMPOSIUM SYMPOSIUM | 99 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

The research is predominantly based on the method of legal dog- of law. Compared to the legal commentaries, the project intends to matics. This method is particularly suitable for the first objective of provide a broader reflection, e.g. as it reviews the basic concepts of solving legal problems. As far as the second objective, namely im- Self-regulation and takes into account the impacts of the interpreta- proving the effectiveness of the instruments is concerned, there- tion. As far as other monographs are concerned, the research will be search will also take into account the experiences made with former one of the first ones directly concerning the GDPR, since it has only interpretation practices. been applicable since 25 May 2018. Moreover, the focus lies on the In concrete terms, this research approach includes an evaluation of administrative implementation and its results-oriented interpreta- Self-regulation in European the current literature, the relevant case-law and the supervisory au- tion rather than the general benefits of Self-regulation. thorities‘ publications as basis for an own interpretation. It is also Data Protection Law planned to interview German supervisory authorities in order to en- All in all, the project refers to these works as a basis, but differs from 16 sure the practical suitability of the interpretation. them by broadening respectively modifying the research focus. The research can refer to some legal commentaries3 as well as mon- Verena Süß ographs4 examining the concept of Self-regulation in various fields University of Bayreuth PRELIMINARY RESULTS [email protected] Prof. Dr. Heinrich Amadeus Wolff So far, the focus has been on the theoretical foundations necessary The research also identified the connecting factors (e.g. sovereign for a solid examination of the GDPR. This includes a definition of decisions affected by previous private activities) and provided more Self-regulation and the correct legal criteria to classify all relevant concrete specifications of constitutional obligations. INTRODUCTION private activities and its reception by the EU or the Member States.

To date, a growing number of voices have been calling for the im- As to 25 May 2018, European Data Protection Law was significantly plementation of “Self-regulation“ into Data Protection Law. Accord- amended in the General Data Protection Regulation (GDPR). And in ing to this concept, private individuals are to be induced to design fact, the new law contains several instruments that can be regarded PLANNED NEXT STEPS their own industry- or business-related data protection regula- as institutions of Self-regulation at first sight, in particular: Codes of tions and to enforce them by means of private control bodies.1 It conduct (Art. 40 f. GDPR), Certifications (Art. 42 f. GDPR) and Bind- The next step will be to apply these theoretical foundations to the GDPR. The objective is to make a proposal for a practical, constitu- is hoped that they will find perfectly adjusted and highly accepted ing corporate rules (Art. 47 GDPR). The question arises, whether European Data Protection Law. It will firstly be necessary to examine tional and results-oriented interpretation of Codes of Conduct, Certi- solutions and therefore contribute to eliminate the shortcomings these instruments exploit the potentials of Self-regulation and are the instruments under former Data Protection Law. On that basis, fications and Binding corporate rules. of traditional Data Protection Law. able to improve European Data Protection Law. the research will finally focus on the current implementation in the

PROBLEM STATEMENT AND RESEARCH QUESTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

Hence the first objective of the project is to examine these instru- since they may want to take advantage of the opportunities arising Significant difficulties are to be expected for the provisions that ex- en “due regard“ (e.g. in Art. 24 (3), 83 (2) GDPR). The content of these ments from a dogmatic legal standpoint. The second objective is to from these institutions (e.g. clarification of indeterminate legal con- plicitly refer to the “adherence to approved codes of conduct“ or to provisions is still unclear. It would therefore be highly interesting to improve their effectiveness by interpreting the relevant provisions in a cepts under the GDPR, avoidance of administrative fines, creation “approved certification mechanisms“ stating that they “may be used seek the view of the supervisory authorities regarding this issue. results-oriented manner. of appropriate safeguards to be allowed to transfer personal data to as an element by which to demonstrate compliance“ or shall be giv- third countries without an adequate level of protection). Although first elements of these instruments already existed under former Data Protection Law and therefore have already received at- Against this background, the key research questions are: What are REFERENCES tention in legal literature,2 there are at least two valid reasons to com- the ideas underpinning the concept of Self-regulation, particularly prehensively examine their current implementation in the GDPR: First- concerning European Data Protection Law? How can the design and [1] Schmidt-Preuß, Matthias, Verwaltung und Verwaltungsrecht zwischen gesellschaftlicher edition, Munich, 2018. ly, the relevant provisions have gained in complexity and raise more enforcement of private provisions be classified in legal terms? In what Selbstregulierung und staatlicher Steuerung VVDStRL 56 (1997), 160–227; Di Fabio, [4] Faber, Angela, Gesellschaftliche Selbstregulierungssysteme im Umweltrecht – unter be- complex legal issues now (e.g. do Certification bodies (Art. 43 GDPR) way are the results of these private actions relevant to the enforce- Udo, Verwaltung und Verwaltungsrecht zwischen gesellschaftlicher Selbstregulierung sonderer Berücksichtigung der Selbstverpflichtungen, Cologne, 2001; Thoma, Anselm und staatlicher Steuerung VVDStRL 56 (1997), 235–277. Christian, Regulierte Selbstregulierung im Ordnungsverwaltungsrecht, Berlin, 2008; belong to the public or the private sector? Do Codes of Conduct (Art. ment of the GDPR? Do the supervisory authorities have a duty to su- [2] Büllesbach, Achim, Transnationalität und Datenschutz. Die Verbindlichkeit von Stockhaus, Heidi, Regulierte Selbstregulierung im europäischen Chemikalienrecht. 40 GDPR) create legally binding obligations relevant to the enforce- pervise these private activities? What are the conditions under which Unternehmensregelungen, Baden-Baden, 2008; Talidou, Zoi, Regulierte Selbstreguli- Eine Untersuchung der kontrollierten Eigenverantwortung für den Schutz der Umwelt ment of the GDPR?). The solution of these problems will contribute these instruments will have more practical relevance and provide erung im Bereich des Datenschutzes, Frankfurt am Main, 2005; Voskamp, Friederike, unter der REACH-Verordnung, Tübingen, 2015. Transnationaler Datenschutz. Globale Datenschutzstandards durch Selbstregulierung, to a better understanding of European Data Protection Law as well as the envisaged results? How are the instruments implemented in the Baden-Baden, 2015. other fields of law containing comparable instruments. GDPR? [3] Beck’scher Online-Kommentar Datenschutzrecht, published by Wolff, Heinrich Amade- us/Brink, Stefan, 24th edition, status as of May 2018; Kühling, Jürgen/Buchner, Benedikt (eds.), Datenschutz-Grundverordnung, Bundesdatenschutzgesetz. Kommentar, 2nd Secondly, these instruments did not have great practical relevance in The project aims to clarify all legal questions raised and to make a pro- the past (except for Binding corporate rules). It is therefore necessary posal for a practical, constitutional and results-oriented interpretation to track the mistakes made in the past (e.g. high costs, lengthy proof the relevant institutions. Its results can contribute to the formation cedures, too few incentives), in order to fully exploit the potentials of of an administrative system of Self-regulation. Furthermore, compa- Self-regulation. nies may use the results as a source of information when participating in one of the examined self-regulating processes. The results will certainly be interesting from the law science point of view. Moreover, companies will indirectly benefit from the project,

100 | SYMPOSIUM SYMPOSIUM | 101 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

As the focus of the thesis lies more on theory building and creating will be interviews and in-depth case studies with companies of dif- typologies and less on testing of existing theories, the research will ferent sizes, but with an emphasis on manufacturers of industrial follow a qualitative, exploratory approach. The main methods used machines and the actors in their ecosystem.

Inter-organizational business model PRELIMINARY RESULTS interdependencies within the context Discussions and workshops with scholars as well as managers of business model interdependencies. Furthermore, the existence of of digitalization small and large manufacturing companies have confirmed the these interdependencies has been demonstrated and discussed present focus on the technical aspects of the digital transforma- based on several cases. 17 tion as well as the lack of research concerning inter-organizational Tobias Pauli Chair of Digital Industrial Service Systems, School of Business and Economics, FAU Erlangen-Nürnberg PLANNED NEXT STEPS [email protected] · is.rw.fau.de Prof. Dr. Martin Matzner The next step in the project is to conduct a series of preliminary and reviews of the existing literature. For a dissertation, of course, interviews as well as a systematic literature review to establish a meaningful and convincing data needs to be collected and analyz- first typology of possible interdependencies, which will then be ed with different theoretical lenses in order to draw relevant and refined and evaluated in a series of further interviews as well as compelling conclusions. INTRODUCTION case studies. So far, the project is only an idea based on discussions

Business models have drawn significant attention from research Internet to find new “ways of doing business”. The current wave of OPEN ISSUES AND POSSIBILITIES FOR COOPERATION and practice for about two decades now, primarily starting with the digitalization is again leading to a great number of changes in com- emergence of a multitude of new e-commerce companies in the panies’ value offerings as well as the way they create and capture late 1990s and early 2000s (Wirtz et al., 2016). In this “new econo- value. Therefore, the business model concept may once more be Other researchers could offer valuable input concerning other cies between companies. my”, the concept of business models offered an approach to ana- useful to analyze the changes. strands of literature and their insights on increasing interdependen- lyze the ways in which the newly founded organizations used the REFERENCES PROBLEM STATEMENT AND RESEARCH QUESTION [1] Bauernhansl, T., Emmrich, V., Paulus-Rohmer, D., Döbele, M., Schatz, A., Weskamp, [5] Leyer, M., Tate, M., Kowalkiewicz, M., Rosemann, M. Bringing Upcoming Technologies M. Geschäftsmodell-Innovation durch Industrie 4.0: Chancen und Risiken für den to a Service Life: Assessing Required Provider and Customer Capabilities. Journal of Initially, many companies start their way towards digitalization by Interdependencies in general, and especially between activities, have Machinen- und Anlagenbau. München. 2015 Service Management Research, 2(1) 42-59. 2018 [2] Benkenstein, M., Bruhn, M., Büttgen, M., Hipp, C., Matzner, M., Nerdinger, F. W. Topics [6] Markides, C. C. Business Model Innovation: What can the Ambidexterity Literature focusing on the technological possibilities and addressing the tech- been proposed before as a central aspect of business models (Zott for Service Management Research – A European Perspective. Journal of Service Man- teach us?. The Academy of Management Perspectives, 27(4) 313-323. 2013 nological challenges that come with it (Leyer et al., 2018). Especially & Amit, 2010). Nevertheless, there seems to be a lack of literature agement Research, 1(1) 4-21. 2017 [7] Wirtz, B. W., Pistoia, A., Ullrich, S., Göttel, V. Business Models: Origin, Development and manufacturing companies moving towards Industry 4.0, driven by an analyzing these interdependencies or those of other business model [3] Beverungen, D., Müller, O., Matzner, M., Mendling, J., vom Brocke, J. Conceptualizing Future Research. Long Range Planning, 49(1) 36-54. 2016 smart service systems. Electronic Markets. 2017 [8] Zott, C., Amit, R. Business Model Design: an activity system perspective. Long Range engineering mindset, concentrate for the most part on the machine elements in more detail. The few papers that do so have a prevailing [4] Gordijn, J., Osterwalder, A., Pigneur, Y. Comparing Two Business Model Ontologies for Planning, 43(2-3) 216-226. 2010 itself and its capabilities (Bauernhansl et al., 2015). To be fair, in a sec- intra-organizational focus and either deal with the interdependence Designing e-Business Models and Value Constellations. BLED Proceedings. 2005 ond step, the possibilities of new services and opportunities as well of elements of a single business model or several business models of as necessities of changes in the own business model are often pon- a single organization in the context of ambidexterity (Markides, 2013). dered. However, the impact of their introduction of smart machines or cyber-physical systems, smart services and their business model Broadly used business model frameworks, tools or ontologies such as changes on other actors and especially their customers has not been the “Business Model Canvas” and “e³ value” also take into account or considered in most cases. even stress the aspect of value creation in a network of actors. How- ever, they do not focus on interdependencies of these actors’ busi- If a machine builder, for example, starts selling smart machines to its ness models beyond a primarily static consideration of the exchange customers, which can gather information about their environment, of value, resources and activities between them (Gordijn et al., 2005). store and analyze it and then maybe autonomously react to it, it This discrepancy between the rising inter-organizational interdepend- seems obvious that this can offer the opportunity for new services on encies of business models as a result of the increasing prevalence of the machine builder’s side (Beverungen et al., 2017). It also seems ev- cyber-physical systems as well as a lack of research in this area calls ident that smart machines and services can improve process efficien- for an intensified consideration of the topic (Benkenstein et al., 2017). cy on the customers’ side. An aspect, however, that has not received too much attention until now is that the new functionalities offered The goal of the thesis, therefore, is to shed light on the interdepend- by the smart machine might be able to initiate further changes in the encies between the business models of companies, which are a part customers’ business model as well, e.g. by enabling them to move of supply chains as well as business and service ecosystems, espe- their value proposition from selling standardized products to highly cially with regard to the capabilities of cyber physical systems. Two individualized offerings. In many cases, the customers also need to of the key research questions arising from this goal are “Which types change their business model and develop new capabilities to fully re- of business model interdependencies exist in business ecosystems?” alize the potential of the new machine and connected services (Leyer and “How does the introduction of cyber-physical systems and smart et al., 2018). This, of course, can trigger changes in other elements of services influence these interdependencies?”. These insights will on the business model or even have a reciprocal impact on the machine the one hand help advancing business model research as a discipline builder. The example demonstrates that the importance of analyzing and on the other hand offer guidance beyond technological aspects such business model interdependencies is a necessity in the times of to managers who are facing the challenge of digital transformation. digitalization.

102 | SYMPOSIUM SYMPOSIUM | 103 ABSTRACTS dations from Webster and Watson (2002). [9] The focus of the consid- overlap. To evaluate the research outcomes, a close cooperation with SYMPOSIUM eration is, in particular, the topics “Approaches to Process Mining in industry partners is essential. In this context, several research projects the production” and “Production management methods to optimize are either in execution or in an advanced planning stage. Within these business processes”. Despite being similar, these different views on projects, a prototypical implementation in at least one business envi- the issue are necessary, as Process Mining and Production Manage- ronment is planned. In addition, I prepare to conduct manager sur- ment are located in different scientific communities with little to no veys to raise the managerial requirements.

RELATED WORK Optimization of production processes A first good approach to optimize manufacturing processes based technological perspective, the integration of production data is very using Process Mining on Data Mining was published by Gröger 2015. [3] He developed interesting. However, the authors miss addressing usability by dean Advanced-Manufacturing-Analytics-Plattform (AdMA) as a mod- scribing interdependencies with conventional production manage- 18 el for gathering, analyzing and optimizing production processes ment methods and technologies. Furthermore, the model does not using Data Mining techniques. The approach is good and from a consider Process Mining. Emanuel Marx PRELIMINARY RESULTS Friedrich-Alexander-Universität Erlangen-Nürnberg, Chair of Digital Industrial Service Systems

[email protected] Being still at the beginning of my dissertation project, so far I was I could verify the relevance of the topic. The next finding expected Prof. Dr. Martin Matzner able to narrow down my research question. By discussing the idea in the nearer future is an analysis of relevant literature to identify with other researchers, supervisor, and various company contacts, specific gaps.

INTRODUCTION

To be competitive and successful, companies need a unique selling long to the best known. Still, the rising complexity of product and PLANNED NEXT STEPS proposition. This can be a product of high quality, low selling price production change the requirements and call for new production or increasingly of customizability. To maintain operability, produc- management methods. With a high volume of available data and The next step will be to conduct a systematic literature review on For an adequate evaluation of the management method, I will ap- tion management is a key factor for companies. In the last years, increasingly powerful IT systems, new possibilities have emerged, Production Management. The objective is to understand the impact ply for a research project together with one or more manufactur- a selection of management methods was developed to optimize opening new fields of research. of production strategies on the production process and to deter- ing companies that declared their interest in the topic. During this the production, from which Lean Management or Six Sigma be- mine the necessary information to be displayed to the process project, the technological part of Process Mining shall be developed analyst to properly optimize relevant processes. As a result, this and implemented in companies and guidelines for change manage- step should generate a list of requirements, which is essential to ment shall be defined. determine the necessary data and develop a Process Mining man- PROBLEM STATEMENT AND RESEARCH QUESTION agement method specifically for the production.

Rising competition and a trend to customizable products lead to of data presentation and decision-making. By solving these issues, shorter and more frequent change cycles and an increasing com- manufacturing companies are better capable of optimizing their plexity of modern production systems. [1] However, due to limits in processes in complex systems and producing more cost and time OPEN ISSUES AND POSSIBILITIES FOR COOPERATION the ability of humans in understanding complex systems, conven- efficiently. Besides, providers of process mining software gain an tional methods to analyze and optimize production processes have additional target audience and are able to expand into new mar- In the current state, my project still needs reliable industry partners to contacts within the Process Mining community would be a valuable come to a limit. [2] Because of this, data-driven approaches gain kets. verify the outcome and to discuss the requirements for practice. Fur- input. Overall, I would be delighted to hear about each new contact importance in planning production processes to extend the capa- thermore, as I rather come from production planning direction, new from science and industry for interesting discussions. bilities of common management and experience-driven analytics. Therefore, the objective of my research is to examine the potential [3] One possibility, which has proven itself capable of optimizing of Process Mining technologies to analyze and optimize production business processes in complex systems, is Process Mining. [4] So processes and to design a concept for enterprises with relevant REFERENCES far, the research of Process Mining in production management has aspects for a successful implementation. The most important re- been scarce. Only a few approaches can handle materials and oth- search questions are: [1] K. Davis, C. Beyond data an analysis. Communications of the ACM, 57(6) 39-41. 2014. Workshops. 99. 2011. er resources alongside the processes where they accrue, whereby • What information does the process owner need from the process [2] Fayyad, U., Stolorz, P. Data mining and KDD: Promise and Challenges. Future Genera- [7] R Hevner, A., T March, S., Park. J. Design Science in Information Systems Research. tion Computer Systems, 13(2-3) 99-115. 1997. MIS Quarterly. 28 (1) 75-105. 2004. these are more generic and hardly useful in the daily practice. [5] mining tools to perform analysis and optimization? [3] Gröger, C. Advanced Manufacturing Analytics. Datengetriebene Optimierung von [8] Peffers, K., Tuunanen, T., A Rothenberger, M., Chatterjee, S. A Design Science Research • Depending on this, which data must be gathered in the produc- Fertigungsprozessen. Dissertation. University Stuttgart. 2015. Methodology for Information Systems Research. Journal of Management Information Making Process Mining understandable and usable for non-ex- tion? [4] Kerremans, M. Market Guide for Process Mining. Gartner, Inc. 2018. Systems. 24 (3) 45-77. 2007. [5] Pika, A. et al. An Extensible Framework for Analysing Resource Behaviour Using Event [9] Webster, J., T. Watson, R. Analyzing the Past to Prepare for the Future: Writing a Litera- perts and offering an operational support are still insufficiently ad- • Are there strong variants in production processes, which prohibit Logs. In: Hutchison. David. et al. Advanced Information Systems Engineering. Interna- ture Review. MIS Quarterly. 26 (2) xiii-xxiii. 2002. dressed challenges. [6] This absence of Process Mining applications a universal concept? tional Conference on Advanced Information Systems Engineering. 2014. in Production Management leads to a lack of management sup- • What is the specific impact of production strategies on processes [6] van der Aalst, W.M.P. et al. Process Mining Manifesto. Business Process Management port and operating instructions. In order to develop a functioning and with this on the result of process mining? system, the requirements of process planer have to be elevated; • What changes must companies make to successfully implement furthermore, data-driven analytics allow new methods and models process mining?

RESEARCH APPROACH AND METHODOLOGY

In order to answer these questions, my research will follow the spec- DSR turns out to be more suitable, as it allows its user to develop busi- ifications of Design Science Research (DSR) by Hevner et al. 2004 [7] ness-relevant and usable artifacts within a scientifically solid frame. using the approach proposed by Peffers et al. 2007 [8]. In contrast to behavioral science, a common research paradigm in information To ensure a solid scientific foundation and to identify relevant litera- systems, DSR focuses on solving concrete problems of the real world. ture, multiple literature reviews are planned, following the recommen-

104 | SYMPOSIUM SYMPOSIUM | 105 ABSTRACTS specific file or calling another process). This greatly simplifies the of safety-critical events. If the normal behaviour is modelled as in [1, SYMPOSIUM supply of instructions for countermeasures for employees who are 3] and discrepancies from this model are taken into account, this is responsible for the security of the company. known as anomaly detection. However, as in the case of [4, 5], lear- Forrest et al. [1] presented one of the first approaches for the detec- ning the behaviour of typical viruses and comparing new sequences tion of safety-critical events through the analysis of system calls back of system calls for similarity is called misuse detection. in 1996. The authors map normal behaviour of different programs While various approaches on both, anomaly and misuse detection, to a database. Known sequences were recorded and split over time have been published over the years, anomaly detection still suffers windows of a fixed number of consecutive system calls. Within this from a high false alarm rate. This is due to the fact that not every Self-adapting Online Analysis of time window, the relative positions of the system calls to each other discrepancy from normal behaviour is necessarily due to a safety-cri- were determined and stored in the database as normal behaviour. tical event. The introduction and calling of new processes can also Host-based Security-critical Events When new unknown sequences arrived, a comparison was made lead to alarms in fast-moving IT systems. Analysing IDS results with with the time windows stored in the database in order to detect many false alarms is very time consuming and can only be done by a in SME Company Networks anomalies in the behaviour of the programs. Over the years, many domain expert. Hence it is very expensive. Consequently, the major other research groups have dealt with the analysis of system calls. In goal when working with anomaly detection methods is to reduce the 19 1999, Warrender et al. [2] performed a comparison of different met- false alarm rate, e.g. by continuously adapting the system. Misuse hods for analysing system calls. In the last few years, some approa- detection does generate less false positives that need time-consu- ches from text mining have also proven to be helpful for analysing ming post-processing by an expert, thus this method was able to as- Sarah Wunderlich system calls. Creech et al. [3] present an approach using semantic sert itself commercially. However, misuse detection, unlike anomaly analysis. Kolosjaji et al. [4] combine convolutional neural networks detection, cannot detect new attack strategies. This is why this work Coburg University of Applied Sciences and Arts known for image processing with recurrent neural networks mainly aims at improving anomaly detection with a lower false positive rate, [email protected] used in text mining. Athiwaratkun et al. [5] take a similar approach. e.g. through the use of domain knowledge. Prof. Dr. Andreas Hotho · Prof Dr. Dieter Landes In general, two main approaches can be distinguished in the analysis

PRELIMINARY RESULTS INTRODUCTION So far, system calls were identified as suitable data source for intru- have been achieved using different kinds of neural networks on the In the age of digitisation, information security is crucial for the long- intrusion detection systems (IDS) are often not used in small and sion detection systems which also have the possibility to give further ADFA-LD data set [6] and will be published shortly. term success of companies. Due to the ever-increasing complexity medium-sized enterprises (SMEs) due to a lack of know-how. The information about the origin of the attack. Good preliminary results of their IT systems and the resulting large amounts of data, the de- PhD project tries to overcome this problem with a self-learning and mands on IT security are also growing. IT security increasingly uses user-friendly system that collects and analyses relevant host-based data mining and machine learning for a comprehensive analysis of data in order to effectively detect and confine attacks. security-related data. However, complex security solutions such as PLANNED NEXT STEPS

PROBLEM STATEMENT AND RESEARCH QUESTION Intended next steps of this work are a refinement and extension of anomaly detection algorithms are in need of a lot of data, supple- the algorithms tested so far and their application to additional data mentary work should be done, to create more system call based data sets. To the best of our knowledge, only three data sets based on sets. While large companies may have sufficient human and financial re- to its origin. Summarising, the doctoral project addresses three key system calls are currently available, two of which are outdated. Since sources to allow security experts to analyse the alerts of intrusion aspects. detection systems, this is usually not possible for SMEs. In order to 1. Identification of suitable data sources. focus on SMEs, it is therefore desirable to present the results in an 2. Analysis, combination, and refinement of various machine lear- easily understandable format with corresponding instructions for ning methods with respect to their applicability to host-based in- OPEN ISSUES AND POSSIBILITIES FOR COOPERATION counteraction. This objective must already be taken into account trusion detection. when developing a new analysis concept. In addition, the data 3. User-friendly and understandable presentation of detected an- The training and validation of anomaly-based systems proves to be data sets for HIDS have been published. Their provision, however, is sources to be analysed should have sufficient information regar- omalies. difficult because labelled data sets are either outdated or cannot critical for the development and evaluation of new methods. Thus, ding an identified security-critical event to be able to trace it back be published for privacy reasons. Additionally, the use of real sys- collaboration on the conception and recording of a new dataset is tem calls is problematic since no ground truth (labels) is available conceivable. and manual labelling would be extremely laborious. As a result, few RESEARCH APPROACH AND METHODOLOGY

Intrusion detection systems can generally be divided into host-based conventional log files. Modern x86/x64 systems do not grant user REFERENCES (HIDS) and network-based (NIDS) systems. While network-based applications direct access to system resources. Applications must IDS usually gather and analyse the traffic in a company network send access requests, e.g. to memory pages or I/O addresses, to [1] Forrest, S. et al.: A sense of self for unix processes. IEEE Symposium on Security and [4] Kolosnjaji B. et al.: Deep Learning for classification of malware system call sequences. based on readily available network components, host-based sys- the kernel via system calls. Consequently, any change to the sys- Privacy. 1996. 120-128. Australasian Joint Conference on Artificial Intelligence. Springer, Cham. 2016. 137-149 [2] Warrender, C. et al.: Detecting Intrusions Using System Calls: Alternative Data Models. [5] Athiwaratkun B. et al.: Malware classification with LSTM and GRU language models tems use information sources of the own operating system they tem – at least in theory – leaves traces in system calls used by the IEEE Symposium on Security and Privacy. 1999. 133–145 and a character-level CNN. IEEE International Conference on Acoustics, Speech and run on or of its processes. While NIDS give a better overview of the applications. [3] Creech G. et al.: A semantic approach to host-based intrusion detection systems using Signal Processing. 2017. 2482-2486 contiguous and discontiguous system call patterns. IEEE Transactions on Computers. [6] Creech, G. et al.: Generation of a new IDS Test Dataset. Time to retire the KDD Collec- overall activity within a company network, HIDS have a potential A significant advantage of system calls over standard log files is the 2014. 807-819. tion. IEEE Wireless Communications and Networking Conference. 2013. 4487-4492 for deeper insights into the operating systems and processes and ability to record and process system calls during their execution, thus into the vulnerabilities used by an attacker. Hence, in order to while log files must first be checked for updates. Conventional log be able to provide countermeasures for exploited vulnerabilities or files therefore can be modified or deleted by attackers before the explain the method of intrusion, HIDS are the preferred solution analysis by an IDS. Another disadvantage is that log files usually do fitted to SMEs. not contain any raw data about the state of a service or process, but Data source and data quality are crucial factors in the use of ma- only information that has already been interpreted. This is also ref- chine learning methods and there are a lot of data sources of dif- lected in the typical free text format of log files, which also makes ferent quality available when looking at computer systems within a analysis by an IDS more difficult. company network. Many programs and services are logging their Since system calls can be recorded depending on the running pro- activities, warnings and errors in some way. However in terms of cess, it is also possible to trace safety-critical events back to the trig- data sources, this work initially focuses on system calls rather than gering process and the planned safety-critical action (e.g. writing a

106 | SYMPOSIUM SYMPOSIUM | 107 ABSTRACTS form. On the other side, to tackle our dynamic scene understand- is inspired by other commonly known large-scale indoor datasets SYMPOSIUM ing problem with a supervised deep learning approach requires a such as ScanNet [2, 3] and planned to be larger, higher quality, with huge amount of data, that is currently not available in the research greater variance and more realistic in terms of the setting and as- domain. Also, comparing different dynamic scene understanding sumptions. We believe this is a main requirement to ensure gener- techniques on newly introduced problems requires high-quality se- alizability. The dataset, as well as proposed algorithms use mobile quences for testing. However, there is currently no large-scale da- devices (Google Tango) with corresponding mobile sensors (IMU, taset available for these specific problems. This makes it a very in- RGB-D). This gives flexibility to choose alternative data formats (e.g. teresting research direction. Therefore, major focus of our research 3D point clouds) [3,4]. Together with the dataset, baseline methods is the implementation and optimization of an acquisition and an- in form of classical approaches are introduced. notation pipeline for efficient 3D recording and labelling. This work Dynamic 3D Scene Understanding 20 PRELIMINARY RESULTS As a first step, we implemented a method for static semantic seg- but worse runtime, requiring a powerful GPU. However, with con- mentation that was efficient enough to run on a mobile device [1]. tinuous improvement of the available hardware in embedded sys- Johanna Wald We then continued with a deep learning based approach for se- tems we expect many deep networks to soon be portable to e.g. Technical University of Munich, Computer Aided Medical Procedures and AR mantic segmentation, part segmentation and 3D captioning [6]. It mobile devices. Therefore, we see the real-time requirement as a processes point clouds in a large-scale setting, memory efficient soft constraint and instead focus rather on more theoretical re- [email protected] enough to process scans of entire apartments all at once. While still search problems such as 3D change detection and 3D optical flow, Prof. Dr. Nassir Navab assuming a static world, [6] gives better quantitative performance, because there, we see a lot of interesting open research questions.

INTRODUCTION PLANNED NEXT STEPS “Internet of Things, a major component of the modern digital transfor- termed 3D scene understanding since it is well-known that under- mation of business and society, potentially has the power to change standing outdoor environments is important for e. g. autonomous Planned immediate next steps involve finishing a) the dataset acqui- it simple in the beginning, the network is trained on semi-synthetic our every-day-life”. These and similar predictions are built on the driving, while processing and understanding indoor environments sition, followed by b) its annotation. Since some parts of the anno- sequences to later be transferred to real world data. Then, for better assumption that the connected devices are intelligent. Therefore, - on the other side - has applications in robotics, augmented reali- tation pipeline are still incomplete, this requires more implemen- comparison, classical baseline methods need to be implemented. machine intelligence became one of the most active research field in ty, entertainment and for assistive technologies such as household tation. Furthermore, the development of a working deep learning For the dissertation, more aspects of dynamic scene understanding, academia and industry. A main sub-topic of it is visual perception robotics. This 3D understanding in static as well as dynamic indoor based method on this dataset requires further exploration. To keep based on the acquired data, are planned to be covered. often referred as computer vision. Computer vision is especially im- environments gives these devices a global understanding of the portant, when effective and intuitive user interaction is required. space around them and, with this, enables spatial reasoning and It gives the devices an understanding of the environment around global planning in aforementioned applications. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION them as the ability to see and process visual input to acquire a high-level semantic map of our highly dynamic world. This is often One open question refers to the size of the dataset that is currently is reliable in terms of correctness of the statistical evaluation? I am built: How much data is really required for testing but also training a also open for discussions about applying deep learning in different machine learning algorithm? This is specifically important when un- modalities since I believe there are many similarities and possibility PROBLEM STATEMENT AND RESEARCH QUESTION supervised learning is applied. Another related, but potentially more for useful exchange. general and scientific, question: When can we ensure our evaluation Our work focuses on algorithms to process real world data of dy- be easily applicable to the real world. This leads back to the challenge namic 3D indoor environments. This covers 3D reconstruction and to be adaptable to dynamics that occur frequently in real world set- scene understanding algorithms with the goal of equipping mobile tings. To achieve this, we want to further investigate how to effective- REFERENCES devices and robots with full spatial and semantic awareness, possibly ly and efficiently use only data from mobile sensors. A major chal-

meeting real-time requirements. The objective of these algorithms is lenge there, is the combination of performance requirements and [1] J. Wald, K. Tateno, J. Sturm, N. Navab, F. Tombari, Real-Time Fully Incremental Scene [4] C.R. Qi, H. Su, K. Mo, L. Guibas, PointNet: Deep Learning on Point Sets for 3D Classifi- to extract semantic information from their surroundings and to au- available resources on those target platforms such as mobile phones Understanding on Mobile Platforms, IEEE Robotics and Automation Letters (presented cation and Segmentation, International Conference on Computer Vision and Pattern tomatically determine properties of the objects therein such as pose and embedded systems. Due to this efficiency aspect, an interesting at IROS), Madrid, 2018 Recognition (CVPR), 2017 [2] A. Dai, A. Chang, M. Savva, M. Halber, T. Funkhouser, M. Niessner, ScanNet: Richly-an- [5] C.R. Qi, L. Yi, H. Su, L. Guibas, PointNet++: Deep Hierarchical Feature Learning on Point and category as well as their spatial, temporal and semantic rela- research direction to be explored is the use of different data formats notated 3D Reconstructions of Indoor Scenes, International Conference on Computer Sets in a Metric Space, International Conference on Neural Information Processing tionships. In contrast to existing approaches, we do not only want to such as raw 3D point clouds, 2D images or 3D data in regular vox- Vision and Pattern Recognition (CVPR), 2017 Systems, 2017 [3] A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Niessner, M. Savva, S. Song, A. Zeng, Y. [6] D. Rethage, J. Wald, J. Sturm, N. Navab, F. Tombari Fully-Convolutional Point Networks get an understanding of static objects but also be aware of dynamic el grids. This involves how to process, store and analyse this data Zhang, Matterport3D: Learning from RGB-D Data in Indoor Environments, Internation- for Large-Scale Point Clouds, European Conference on Computer Vision (ECCV), changes. Many state-of-the-art methods operating indoors are not for dynamic 3D scenes understanding. We hope that our findings al Conference on 3D Vision (3DV), 2017 Munich, 2018 adaptable to those changes [2, 4, 5], even though there is a high de- are helpful for the research community to shape future work and mand for such methods from academia and industry. An important research directions, enabling contributions in a field that has not yet but related aspect of this research is that methods are expected to been explored much in the research domain.

RESEARCH APPROACH AND METHODOLOGY

Since a main requirement of the project is to work in the real world, mental Scene Understanding on Mobile Platforms [1] has for example the research and the methodology are applied research. Methods been deployed to Google Tango phones to a) show its advantages used in this project are placed in the computer vision and computer in terms of performance – being the first semantic-segmentation graphics domain, often taking advantage of the power of machine framework to run in real-time on-board of a mobile device – while learning algorithms such as deep learning. To ensure its generaliza- b) sill giving comparable performance to other state-of-the-art bility, live demos are usually implemented. These act as interactive methods. Publicly available datasets and corresponding matrices tests and potentially help to find weaknesses of the algorithms to are used to evaluate against related work. This ensures a fair com- help scale up to real world settings. Our work Real-Time Fully Incre- parison that gives a good understanding on how our methods per-

108 | SYMPOSIUM SYMPOSIUM | 109 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

To increase the presence induced in the user, we first need to ex- methods that induce higher presence in the user. After generating plore current interaction methods in virtual reality and then expose an approach, we check the feasibility and possible physical con- factors that influence presence positively or negatively. After sin- straints. After reifying the concept we build a working prototype gling out an aspect, e.g. how heavy the user expects a virtual object that lets users immerse into and interact with the virtual world. To should be, we examine related research [3,4,5], generate ideas and evaluate the effect of such an approach we conduct user studies in concepts on how to solve an issue, improve current concepts or order to empirically evaluate the approach and find out possible Haptic Feedback and Interaction develop new approaches on how to allow interaction and feedback effects and issues that need to be improved. 21 in Virtual Reality PRELIMINARY RESULTS As this is my first year in my PhD most projects are still being devel- (e.g. drones) often does not fulfil the needs for research projects. oped and evaluated. Therefore we are in an exploratory phase and Specification of commercially available drones are not suited for Matthias Hoppe develop various prototypes for testing. mounting of additional hardware and are not accurately controlla- ble by automation and therefore limit the achievable effects that is LMU Munich/Human-Centered Ubiquitous Media So far we utilize the fact that simple illusions can often be used to needed to create haptic feedback. [email protected] trick the user into feeling a higher presence. Purchasable hardware Prof. Dr. Albrecht Schmidt

PLANNED NEXT STEPS INTRODUCTION We are planning to further investigate methods of providing simple Currently, we are looking for expert users for evaluating instructions Virtual Reality allows users to move to different places, he or she accurate interaction methods nor do they provide more representa- feedback cues that give the illusion of feeling feedback and allowing for building custom drones and a Unity3D interface that allows the can immerse into an underwater world, outer space or even fictio- tive feedback than simple vibration signals. the users to interact with the virtual world in a more direct way that automation of these drones. nal worlds that withstand physics or logic. This is often accomplis- does not involve input methods such as hand-held controllers. hed by different modalities. Such as head-mounted displays (HMDs), Haptic exploration consists of two main components, first the ab- Further, the planing, design and construction of several prototypes for changing what the user sees. Headphones that let the user hear ility to interact with the surroundings (e.g touching objects, having By using VIVE Trackers we are going to introducing haptic drones into for providing haptic feedback in virtual reality are being developed. sounds and music and drown out the sounds from the real world. a physical impact) and second, the reaction of the world by giving the virtual space by using the lighthouse system. This eliminates the But also, hand-held controllers, that let the user interact and feel adequate feedback and reacting to the users actions. need for external tracking, such as OptiTrack. simple feedback signals of the virtual world. The quality and accuracy of interaction methods are commonly rea- The quality and fidelity of video and audio increases as technology lised with some form of controller. These hand-held devices allow OPEN ISSUES AND POSSIBILITIES FOR COOPERATION improves. Higher resolution displays remove the "flyscreen" effect, interacting with the virtual world by pressing buttons as a form of where single pixels and the black space in between are visible. High "grabbing" or activation. Additionally, feedback can be felt as vibra- fidelity headphones can create accurate sound representations. Im- tion motors are being activated at the point of contact or impact. As some planned prototypes include complex automation tasks we bots. The robots would need to accurately react to and mimic users proved sound engines can create higher quality three-dimensional These low-resolution feedback methods can serve as a notification are currently looking for a cooperation with people that have expe- movements sounds. The accuracy and speed of tracking improve and therefore and signal of interaction, but are too coarse to make interactions, rience in the field of robotics and automation of KUKA or similar ro- erase some factors of motion sickness. such as touching an object, feel "real".

However, humans do not only rely on their vision and hearing, but Therefore, we are developing and evaluating methods that give the REFERENCES also on their sense of touch. Humans tend to physically explore their illusion of interacting with complex surroundings without the need surroundings by touching objects, feeling the resistance, weight and to exactly rebuilt the virtual world with physical tokens. We rather [1] https://www.leapmotion.com/ [5] Kazuki Nagai, Soma Tanoue, Katsuhito Akahane, and Makoto Sato. 2015. Wearable surface structures. Therefore, ways of providing adequate haptic want to build adaptive, dynamic systems that give the user the il- [2] Slater, Mel. "A note on presence terminology." Presence connect 3.3 (2003): 1-5. 6-DoF wrist haptic device SPIDAR-W. In SIGGRAPH Asia 2015 Haptic Media And Con- [3] J. Blake and H. B. Gurocak, "Haptic Glove With MR Brakes for Virtual Reality," in IEEE/ tents Design. ACM, 19. and tactile feedback are needed to induce a high level of presence, lusion of a completely physical world where touching everything is ASME Transactions on Mechatronics, vol. 14, no. 5, pp. 606-615, Oct. 2009.doi: 10.1109/ that makes the user feel as he or she would be actually in the virtual possible by giving simple impulses. TMECH.2008.2010934 world. Current solutions, e.g. hand-held controllers, neither provide [4] Pedro Lopes, Sijing You, Lung-Pan Cheng, Sebastian Marwecki, and Patrick Baudisch. 2017. Providing Haptics to Walls & Heavy Objects in Virtual Reality by Means of Electri- cal Muscle Stimulation. In Proceedings of the 2017 CHI Conference on Human Factors in Computing Systems. ACM, 1471–1482. PROBLEM STATEMENT AND RESEARCH QUESTION

In contrast to other modalities such as vision and hearing, higher ble. Our approach is to combine both, bringing the user’s hands into resolution hardware or better algorithms are suitable to increase the the virtual world, while still being able to provide haptic feedback. level of immersion. For haptic exploration, new concepts of providing haptic feedback and enable interactions are needed. It is pos- For a seemingly real world in virtual reality, a certain technological sible to replace hand-held controllers with sensors that allow hand level, the so-called immersion, is needed. A higher immersion helps tracking [1]. Such sensors allow to bring the user’s hands into the to induce a higher level of presence in the user. And the higher the virtual world and show a virtual representation of the user's hands. presence, the more the user 1 accepts the virtual surrounding as The user can interact with the virtual world and push or grab objects, "real" and is more engaged in what is presented to him or her [2]. but due to the missing vibration motors, no haptic feedback is possi-

110 | SYMPOSIUM SYMPOSIUM | 111 ABSTRACTS concept for the development of parallel and concurrent real-time is used in order to provide simple and reusable interfaces for the SYMPOSIUM applications. adapted patterns. The distribution of the workloads among various Following the above-mentioned work on parallel real-time sched- processing cores and hardware architectures will be handled by the uling, the research approach is subdivided into four subdomains: underlying software framework. On that account, different sched- At first, widely known parallel and concurrent design patterns, for uling mechanisms will be evaluated so that optimal collaborations example map-reduce or divide-and- conquer, shall be investigated between the patterns and the algorithms lead to reduced compu- for applicability in real-time systems. Secondly, a mathematical ap- tation times. Also, the application of real-time safety patterns, as for proach shall give insight about the worst-case timing behavior with example the heterogeneous redundancy pattern, on the parallel and respect to different amounts of workloads and a variety of schedul- concurrent design patterns will be contemplated so as to improve ing algorithms on a diverse number of processing units and hard- the reliability. At last, benchmarks will then be conducted on the im- Parallel & Concurrent ware architectures. In the third subdomain, the adapted patterns plementation in order to reassure the correctness of the theoretical will be implemented using the C++ programming language which is work under real conditions. The outcome of the four subdomains Real-Time Design Patterns taking important steps towards pattern-based, fine-grained paral- shall be summarized in the aforementioned pattern catalog. 22 lelism and heterogeneous computing. A template-based approach Michael Schmid PRELIMINARY RESULTS

OTH Regensburg, Faculty of Electrical Engineering and Information Technology The work so far has focused on a design pattern from the area of tation has been established. Furthermore, parallel patterns are cur- [email protected] · www.las3.de distributed computing. A concurrency pattern, namely the Active Ob- rently being investigated and their timing behavior is being analyzed Prof. Dr. Jürgen Mottok ject pattern which provides synchronized access to an object shared in order to adapt them for time-sensitive environments. Results on between multiple threads of execution, has been adopted for the their worst-case timing behavior are expected soon, as an adjust- proper utilization in time-critical systems. The response time analysis ment from the works proposed in [2] and [3] is manageable. has been conducted and a simple, reusable C++ template implemen- INTRODUCTION

Nowadays, according to Herb Sutter, one can find a „Heterogeneous since. These add a great benefit to the software development pro- Supercomputer in every Pocket“ [1]. This implies that software devel- cess of massively parallel and heterogeneous hardware by providing PLANNED NEXT STEPS opment needs to undergo a decisive change in order to fully exploit reusable, tested and well-proven development paradigms. the hardware resources at hand. The high performance computing However, those patterns do not necessarily satisfy the strict require- The aforementioned parallel patterns need further investigation Furthermore, practical evaluations on heterogeneous hardware and distributed computing sectors have dealt with parallel and con- ments imposed by real-time systems and therefore are not easily ap- regarding their response times. Subsequently, they shall be imple- architectures of the implemented patterns will follow in the future current computing for some decades. They therefore have accumu- plicable in time- and safety-critical environments, as for example the mented in the C++ programming language in conjunction with dif- in order to review the correctness of the real-time analysis. Finally, lated experience in developing scalable software for heterogeneous automotive or avionic sector. Naturally, these sectors would benefit ferent scheduling algorithms in order to identify optimal collabora- the successfully adapted patterns of the previous steps will then be computer architectures, as for example multi-core processors, field greatly from the existing knowledge. As a consequence, thorough in- tions between the design patterns and the dispatching mechanisms. united in a pattern catalog. programmable gate arrays or digital signal processors. Many design vestigations and adaptions of those patterns, regarding their timing patterns for parallelism and concurrency have been established behavior and safety, must be conducted. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

PROBLEM STATEMENT AND RESEARCH QUESTION Due to the large variety of parallel and concurrent design patterns, tation on real hardware. it is not possible to conduct the proposed research approaches on Furthermore, many perspectives have to be considered in the sec- The development of parallel and concurrent applications is a cum- dition, is often used to analyze the feasibility of the system at hand. all of the patterns that are suitable for a real-time adaption. Also, tor of real-time scheduling which makes a thorough analysis cum- bersome endeavor. Obstacles, as for example improper data local- This thesis is located at the intersection of the aforementioned sec- more ways of analyzing the timing behavior are conceivable. This bersome. Feedback and reviews concerning the investigations of ity, communication overhead or data races, pose a big challenge tors and deals with the investigation of fine-grained parallelism and implies that collaboration on the adaption can be performed on the timing behavior are therefore indispensable. for scientists and programmers. On that account, developers can concurrency patterns in heterogeneous real-time systems. Conse- different levels. As an example, this work would benefit greatly Thus, collaborations with experts from the field of theoretical com- resort to numerous, well studied design patterns which help them quently, common design patterns are identified for a suitable adap- from timing simulations conducted on the execution of the design puter science, especially in the real-time area, would greatly con- to solve repetitive parallel and concurrent problem statements and tion to the real-time sector and their timing behavior on different patterns under the use of the selected scheduling algorithms as an tribute to this thesis. thus, allow them to focus on the implementation of the required computational resources is evaluated. The results will be collect- intermediate step between the timing analysis and the implemen- algorithms. For this purpose, parallel computations are broken into ed in a pattern catalog for easy to handle parallel and concurrent smaller workloads and scheduled among different processing cores real-time patterns that supports software developers in their de- or hardware architectures. Yet, a factor that plays an important role sign of real-time applications on heterogeneous multi-core archi- in time-critical systems is the timing behavior of computations: Re- tectures. In summary, this thesis is concerned with the adoption REFERENCES al-time tasks must meet their deadlines in order to maintain correct of well studied parallelism and concurrency design patterns to the [1] Sutter. Herb. Welcome to the Jungle. Web site. Available at [4] Schmidt. Douglas. Pattern-Oriented Software Architecture, Volume 2: Patterns for system operation. A common way to express the timing behavior of real-time domain and the analysis of their applicability. https://herbsutter.com/welcome-to-the-jungle/. Accessed August 07, 2018. Concurrent and Networked Objects. Wiley. 2000. those tasks is through the worst-case response time, which, in ad- [2] Maia. Cláudio. Response-Time Analysis of Synchronous Parallel Tasks in Multi- [5] Sha. Lui. Priority Inheritance Protocols: An approach to Real-Time Synchronization. processor Systems. Proceedings of the 22nd International Conference on Real-Time IEEE Transactions on Computers, vol. 39(9). 1990. Networks and Systems. 2014. [3] Parri. Andrea. Response time analysis for G-EDF and G-DM scheduling of sporadic DAG-tasks with arbitrary deadline. Proceedings of the 23rd International Conference RESEARCH APPROACH AND METHODOLOGY on Real-Time Networks and Systems. 2015.

Much literature can be found on the scheduling of parallel appli- terns. Some of them are well described in [4]. Yet, a thorough search cations in real-time systems. In [2] and [3] Maia et al. and Parri et of the state of the art literature al. provide means to calculate the worst-case response time and a yielded no real-time investigation of these patterns. Synchronization schedulability analysis of parallel fork/join and directed acyclic graph patterns on the other hand are well examined in the real-time sector, tasks, respectively. Both of their task-based approaches rely on the as for example by Sha et al. in [5], and thus are not an essential part interference that parallel tasks suffer by either themselves or other of this work. tasks in the system. This methodology can be adopted in order to While the previously mentioned works of [2] and [3] focus on sched- realize the timing analysis of the selected parallel patterns. In terms uling of parallel real-time tasks, this work concentrates on the in- of distributed systems, there exist a vast amount of concurrency pat- dividual design patterns with means to provide a “building block”

112 | SYMPOSIUM SYMPOSIUM | 113 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

In a first step, we use quantitative measures, based on graph theory, get a deeper understanding of processes in a specific project, for ex- mathematical models, and predictions of the behavior of single de- ample, why a certain developer uses a certain communication chan- velopers or developer groups, to get an overview of the core organ- nel at a certain point in time. Another interesting approach can be izational principles, and to compare them. For example, a centrality to analyze specific situations in one project and compare the results measure shows us the most active and consequently the most im- for OSS development projects with commercial software projects. portant developers in the OSS project. The distribution of a centrality For this analysis, we are preparing a study at our industry partner, measure characterizes the strength of the dependence of a project Microsoft Research. Human-Oriented Analysis of on single developers. Our approach is rooted in the area of network Our methodological approach is backed by previous research. A analysis and repository mining. Networks are good approaches to research group around Apel and Joblin started using networks to Software Developer Networks model the interaction between different developers, because net- analyze development in open source software projects [2,3]. Our re- works focus on the main aspects of interaction and neglect unneces- search builds on their findings and data operationalization, in that 23 sary details [3]. We use version-control data from repositories, mail- we reuse their tools, e.g. Codeface (available at: https://github.com/ ing lists, and issue trackers from OSS projects to build networks, in siemens/codeface) and their intuitive notion of developer collabora- which we model developers and/or source-code artifacts as vertices tion. Another group of researchers around Häussler and Harnisch Barbara Eckl and interactions as edges. For example, when looking at develop- evaluate the organizational structure of the Linux community and ers’ mail communication, developers are vertices and each mail is how developers advance in the developers’ organizational structure University of Passau, Faculty of Computer Science and Mathematics a directed edge between the sender and the receiver. Then, we use [1]. We also build on this work, as it provides a good starting point to [email protected] · http://www.fim.uni-passau.de/forschung/piccard/team/ network measures to analyze these networks. With mathematical understand specific processes in the most well-known and biggest Dr. Janet Siegmund models, rooted in different disciplines, such as epidemiology [4], we OSS project, Linux [1]. A mathematical approach of organizational model evolution processes in software development projects and structure is described by Ravasz and Barabási [5], who model struc- predict project behavior. ture based on hierarchy. We use their model to build our own math- These models can be evaluated by comparing the observed behavior ematical theory on how networks evolve. INTRODUCTION with the predicted behavior. In a second step, we plan to validate and By combining previous approaches, we gain a novel perspective on complement the quantitative measures by qualitative research to the problem of developer collaboration. Moreover, we combine dif- get a more detailed understanding. Regarding qualitative methods, ferent data sources to get a broad The human factor plays a key role in software development, because organization differs from commercial projects. In this thesis, we fo- we will use surveys and interviews. These methods can be helpful to humans are developers and users of software. Thus, understand- cus on human behavior during the software development process. ing human behavior in development process has been attracting Specifically, we search for structures in developers’ collaboration and researchers’ attention. A high proportion of the software we use is communication to understand organizational structures of OSS pro- open-source software (OSS). This kind of software is particularly in- jects. Our goal is to provide insights and tools to optimize the devel- PRELIMINARY RESULTS teresting for research, as it is mainly developed by volunteers and opment process of OSS. thus the project Software projects usually change over time, for example, in the num- bined data sources, e.g., mailing list and issue information. We have ber of developers, their organizational structure, and the turn-over computed first measures and created first visualizations of network of developers. To get as much information as possible, we have com- characteristics, such as hierarchy. PROBLEM STATEMENT AND RESEARCH QUESTION

Organizational dynamics, which affect failure or success of projects, • To compare projects according to their structure, we need to find and the behavior of individual developers are not well understood. comparable measures, for concepts like hierarchy [5]. What are the PLANNED NEXT STEPS However, to optimize the development process, it is important to organizational structures of the projects? track organizational dynamics. For example, by analyzing the organ- • To diagnose problems, we need to track changes. How does the Our first line of work concentrates on finding an intuitive notion of PhD, in which I have explored numerous ideas and directions. Espe- izational structure of the Linux project researchers found that the organization evolve over time? hierarchy and on the analysis of the dynamics of hierarchy in the cially exploring hierarchy and enriching this understanding by quali- project would profit from a reduction of the communication cost. • To understand how a single developer influences a project, we need context of OSS projects. Then, we plan to evaluate and extend our tative methods is very promising, so I am convinced that this plan will With the growing number of contributers, the maintainers are over- to understand the behavior and characteristics of a single develop- definition with qualitative methods. I am still in the first year of my lead to a successful dissertation. loaded with the increasing bureaucracy1. er. How do they influence the developer’s role in an OSS project? Aligning coordiation needs and efforts, is of particular interest in OSS • To predict the evolution of a role of a developer, we need to under- projects. In such projects, volunteers play a crucial role and organi- stand her motivation. How does a developer choose the parts on zational structures are often very loose [2]. Many OSS projects have which she like to work? OPEN ISSUES AND POSSIBILITIES FOR COOPERATION a huge impact on software engineering nowadays. For example, em- To answer the research questions, we combine approaches from ployees of Google contribute to Linux and build Android on the Linux empirical software engineering, behavioral research, and network A big issue is the reusability of our analyses in companies. It would our results from OSS projects with industrial projects, we would like kernel. Thus, although Linux and Android are OSS projects, they are analysis. For example, we investigate the explanatory value of new be very interesting to evaluate our measures in companies to con- to cooperate with an industry partner. We have a cooperation with also driven by companies. multi-dimensional network models for understanding the develop- trast them with OSS projects, so that we can understand in which Microsoft Research. However, we are seeking contact to further com- Before we can improve the organizational structure, we evaluate ment process. Thus, as a first result, we expect a theory that explains aspects they are different and in which they are equal. To compare panies. how developers communicate and collaborate, and also understand how OSS developers coordinate their work. This can improve guide- their behavior as individuals in the software project. Moreover, de- lines that support developers to communicate efficiently. Further, velopers have different roles and can move up in the hierarchical the results will allow to improve support and lead systems, for ex- REFERENCES structure of an OSS project. ample in GitHub. This way, we improve the organizational flow in the With regard to the described problems, we formulate OSS projects. [1] M. Hanisch, C. Häussler, S. Apel, and S. Berreiter. Developers’ progression from pe- [3] M. Joblin, S. Apel, and W. Mauerer. Evolutionary trends of developer coordination: the following research questions: riphery to core in the linux kernel development project. Proceedings of the Academy a network approach. Empirical Software Engineering, 22(4):2050–2094, 2017. • To optimize communication, we must understand communication 1 https://www.heise.de/newsticker/meldung/Linux-Entwickler- of Management Conference 2018. [4] A. Fall, A. Iggidr, G. Sallet, and J. J. Tewa. Epidemiological models and lyapunov func- [2] M. Joblin, W. Mauerer, S. Apel, J. Siegmund, and D. Riehle. From developer networks tions. Mathematical Modelling of Natural Phenomena, 2(1):62–83, 2007. habits of developers. Which different communication channels do Kernel-Community-wird-unter-eigener-Buerokratie- to verified communities: A fine-grained approach. In International Conference on [5] E. Ravasz and A.-L. Barabási. Hierarchical organization in complex networks. Physical they use in OSS projects, and how do they differ in usage? zusammenbrechen-4030460.html. Last accessed on 26.08.2018 Software Engineering, 563–573. IEEE Computer Society, 2015. Review E, 67(2):026112, 2003.

114 | SYMPOSIUM SYMPOSIUM | 115 ABSTRACTS 2. Methods to measure semantic similarity: This step includes re- termine the reliability from the human perspective and in relation SYMPOSIUM searching established state of the art methods within and outside of to background and experience level. To determine the most suitable the software development domain. Afterward, we select and imple- method for semantic coupling, we compare and interpret the human ment the most promising methods. ratings versus the computed scores per method. In a follow-up sur- 3. Evaluation and answering of research questions: We will conduct vey, participants will assess the perceived usefulness of semantically a survey among developers of different background and experience grouped artifacts by comparing manual groupings with computed level. Each participant rates random code artifacts on a similarity groupings. Constructing a Semantic Map of Source scale. Subsequently, we compare the inter-rater agreement to de- Code Artifacts to Enhance the Software RELATED WORK Development Process Joblin et al [2] investigate the coordination of software developers tic measure better reflects a developer’s mental representation of a 24 based on technical artifacts. Whenever developers have contribut- software system than other coupling measures. Their similarity re- ed to semantically coupled content at the same time, the authors sults are based on Latent Semantic Indexing. assume collaboration between these developers. Our research on Similarity measures that leverage knowledge-based tools, such Tina Schuh collaborative as WordNet, perform poorly in the software domain, because the behavior among software developers is closely related to this work meaning of a word in general-purpose conversations often differs University of Passau, Faculty of Computer Science and Mathematics and our starting point is to replicate and build their results. In the from its use in the software context [7], e.g., ‘disabled’ in general [email protected] long term, semantic coupling will help us to understand how devel- means ‘handicapped’, in the software context, it means ‘not active’ Dr. Janet Siegmund opers coordinate their maintenance efforts. [5]. Researchers have built tools to derive similar terms in the soft- In an empirical study, Bavota et al. [1] have shown that a seman- ware context [4-6], which we intend to use in our work.

INTRODUCTION PRELIMINARY RESULTS

When we decompose a software system, we find an extensive col- perceived by developers only partly [1]. By contrast, coupling based At this early stage of the project (started July 2018), I have familiarized projects, ensuring the proper handling of file encodings, removing of lection of software artifacts that are linked in various ways. Un-in- on semantic information found in identifiers and code comments myself with the topic, key players, and literature. I have also invest- potentially problematic characters (e.g., control characters, emojis), formed changes in one part may unintentionally affect other parts seems to align better with a developer’s mental model, especially ed time in preparing the data: Sifting through sample open source and annotating code artifacts with syntactical information. of the system [2] and lead to bugs. Therefore, in the software de- as it uncovers otherwise hidden relationships [1]. For example, two velopment process, an essential skill is the ability to form an un- individual software artifacts that contain similar terms (e.g., class derstanding of code parts of interest, as well as of their interplay. It VideoPlayer with a method called muteAudio() and class VideoEditor entails the navigation and skimming through source code artifacts that contains a comment “\\turn off audio”) are considered to be —not all of which are relevant. To assist developers, some develop- semantically related. As a final outcome, we envision a map where PLANNED NEXT STEPS er tools (e.g., SciTool’s Understand) offer a code analysis based on semantically related code artifacts are grouped into distinct areas static code features such as function calls, variable access, and in- (similar to Richard Wettel’s CodeCity). To enrich the visulization by The next step of the project is to finalize the selection of semantic heritance relations, to automatically couple associated entities and coordination information, we envision to also integrate in which se- coupling methods and start with the implementation. display them accordingly. This approach is based on structures in mantic neighborhoods others are currently active. the code, and although it is beneficial [1], it reflects the coupling as OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

PROBLEM STATEMENT AND RESEARCH QUESTION The choice of artifact samples and questions for the survey are cru- Ideas addressing the visualization of semantic areas are also more cial to ensure a meaningful and valid outcome. We would appreciate than welcome. When starting to work on unfamiliar code, developers often spend a opers of different background and experience level agree about semantic input regarding the design of our survey and its experimental set-up. considerable amount of time and effort to orient themselves. Assist- similarity with each other and with the computed results? To what extent ing developers in grasping the conceptual dimension of a system will do developers perceive the semantic grouping as useful? not only accelerate this phase, by revealing possibly hidden areas of REFERENCES potential impact, it may also serve as a valuable means to prevent We aim to contribute an evaluation of different semantic similarity bugs and inefficiency. measures in the software domain. Furthermore, by translating se- [1] G. Bavota, B. Dit, R. Oliveto, M. D. Penta, D. Poshyvanyk, and A. D. Lucia. An Empirical [5] J. Yang, L. Tan. SwordNet: Inferring Semantically Related Words from Software Con- mantic closeness into a visual map, we will offer a fresh perspective Study on the Developers' Perception of Software Coupling. In Proceedings of the text. International Conference on Software Engineering, 692-701. (2013) In Empirical Software Engineering, Volume 19 Issue 6, 1856-1886. (2014) Our key research question is: What is the best method to measure se- on software systems. In the future, this information can be com- [2] M. Joblin, S. Apel and W. Mauerer. Evolutionary Trends of Developer Coordination: [6] Y. Tian, D. Lo and J. Lawall. SEWordSim: Software-Specific Word Similarity Database. mantic similarity in the domain of software development? bined with developer activity, e.g., to study whether developers pre- a Network Approach. In Empirical Software Engineering, Volume 22 Issue 4, 2050-2094. In Companion Proceedings of the 36th International Conference on Software Engineering, fer to stay within the (2017) 568-571. (2014) [3] S. Harispe, S. Ranwez, S. Janaqi and J. Montmain. Semantic Similarity from Natural [7] G. Sridhara, E. Hill, L. Pollock and K. Vijay-Shanker. Identifying Word Relations in As no objective measure for similarity exists [3], we are interested in boundaries of specific semantic areas. Incorporated into an inte- Language and Ontology Analysis. Morgan & Claypool Publishers (2015) Software: A Comparative Study of Semantic Similarity Tools. In 16th International how well human assessment and computed similarity align. Thus, grated development environment (IDE) and based on a developer’s [4] M. J. Howard, S. Gupta, L. Pollockand and K. Vijay-Shanker. Automatically Mining Conference on Program Comprehension, 123-132. (2008) to evaluate our operationalization of semantic similarity, we invite current activity, semantic similarity can be used for concrete recom- Software-Based, Semantically-similar Words from Comment-Code Mappings. In Proceedings of the 10th Working Conference on Mining Software Repositories, 377-386. developers of and experience level to rate the appropriateness of mendations to evaluate or modify code elsewhere. (2013) similarity and semantic grouping of artifacts: To what extent do devel-

RESEARCH APPROACH AND METHODOLOGY

1. Data: We use codeface to mine and store data from publicly under different conditions (e.g., inclusion or exclusion of reserved available open source projects. In an initial preprocessing step, keywords from code, but not comments). Further preprocessing we enrich the source code artifacts with syntactical annotations steps include: Splitting the textual data into individual words (to- on the lexeme level (e.g., marking of keywords, identifiers, com- kenization), removal of the most common english words (i.e, ‘stop ments) to enable programming language independent analyses words’) or reducing words to their stem.

116 | SYMPOSIUM SYMPOSIUM | 117 ABSTRACTS agencies that have been proposed by scholars and practitioners ent gain settings (e.g., money, service personalization) is analyzed SYMPOSIUM are evaluated. empircally: Does willingness to disclose data differ under different • The experimental study investigates two related issues. First, us- uncertainty structures? Does the gain setting have an impact on ers’ behavior when choosing a service provider is explored: Do data disclosure? This idea builds on studies on uncertainty in the subjects want transparency? Do content and services providers context of privacy (e.g., Acquisti & Grossklags, 2012), users’ data have an incentive to be transparent? How does transparency af- disclosure behavior (e.g., John, Acquisti, & Loewenstein, 2011) and fect competition? This idea builds on research in the context of the privacy calculus (e.g., Dinev & Hart, 2006). Up to now, privacy uncertainty and decision analysis (e.g., Li, Müller, Wakker, & Wang, research has mostly used scenario-based surveys. Thus, econom- 2017) and is among the first experimental studies to explore am- ic laboratory experiments can help to isolate behavioral effects biguity attitudes in the context of data. Second, users’ data disclo- and investigate privacy in the context of revealed preferences. The Collection of Personal Data: sure behavior under risk, ambiguity and certainty as well as differ- Competitive and Behavioral Effects 25 PRELIMINARY RESULTS

The interdisciplinary review of the economic, information systems ings indicate a subtler relationship that is strongly influenced by Victoria Fast and marketing literature provides a basis to draw overreaching pol- an additional set of factors (e.g, network, scale, scope and effects, icy implications. Foremost, although data generally provides a firm data-induced switching costs), which may mutually influence and University of Passau (Research Group Data Policies) with a short-term competitive advantage, this advantage does not reinforce each other. [email protected] · http://www.datapolicies.net necessarily need to translate into market power. Instead, the find- Dr. Daniel Schnurr · Prof. Dr. Jan Krämer

PLANNED NEXT STEPS INTRODUCTION • Feedback from conference presentations on the literature review • After finalizing the experimental design, experimental sessions • Data is considered to be the new oil of the digital economy and es based on data power lead to market power in the long run and is currently incorporated into the working paper. Policy implica- will be run. the key ingredient for today’s datadriven business models. Firms protect firms from competition is debated controversially in the lit- tions are discussed with policy makers and practitioners. that have access to a large set of unique user data are among the erature and is therefore explored in depth in this thesis. highest valued firms measured by market capitalization and the • As data-based competitive advantages are dependent on a firms’s transaction value of recent mergers and acquistions (e.g., Facebook superior ability to collect and process data, data collection does and Whatsapp) underscores the importance and value of access not only offer benefits to users but also makes them vulnerable. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION to data. Users benefit from the collection of personal data asit In particular, information asymmetries between users and firms provides them with access to (personalized) services, recommen- regarding data collection, usage and disclosure induce uncertainty • I would appreciate feedback from other empirical researchers on how transparency can be implemented in practice. Further issues dations and more relevant advertisements. However, the central and potential harm. For example, firms might use personal data for the experimental set-up and design: e.g., how to collect personal like algorithmic transparency could be explored. Legal expertise role of (personal) data raises several important regulatory concerns price discrimination, spam or data sharing with third parties. Re- data in the lab (or field)? How to incentivize participants? on how to transform policy recommendations into a regulatory and challenges. Thus, the goal of this dissertation is to study the search shows that (Perceived) Privacy Risk, i.e., “the expectation of • Moreover, technical expertise would help to better understand framework would also be helpful. implications of the collection of (personal) data for firms and users. losses associated with the disclosure of personal information” (Xu, Two research projects carried out in the context of this thesis are Dinev, Smith, & Hart, 2011, p.804), may increase Privacy Concerns summarized in the following: a literature review on the competitive (e.g., Dinev & Hart, 2006) and negatively impact users’ disclosure REFERENCES effects of data and an experiment on the behavioral effects of data intentions (e.g., Malhotra, Kim, & Agarwal, 2004). In order to assess collection. privacy risks, users are said to evaluate the probability and severity [1] Acquisti, A., & Grossklags, J. (2012). An Online Survey Experiment on Ambiguity and [5] Li, Z., Müller, J., Wakker, P. P., & Wang, T. V. (2017). The Rich Domain of Ambiguity • As online platforms like Google Search, Amazon and Facebook have of the negative consequences related to data disclosure. In reality, Privacy. Communications & Strategies, 88(4), 19-39. Explored. Management Science. [2] Dinev, T., & Hart, P. (2006). An extended privacy calculus model for e-commerce [6] Malhotra, N. K., Kim, S. S., & Agarwal, J. (2004). Internet users' information privacy gained substantial market shares in various online content mar- however, these two aspects are rarely known to users and users transactions. Information Systems Research, 17(1), 61-80. concerns (IUIPC): The construct, the scale, and a causal model. Information Systems kets, they now frequently represent gatekeepers between internet find themselves in situations of ambiguity (unknown or uncertain [3] European Commission [EC] (2016). Digital Single Market. Retrieved from Research, 15(4), 336-355. users and internet content. This market position allows those firms probabilities) rather than risk (known probabilities). According to https://ec.europa.eu/digital-single-market/en/policies/online-platforms [7] Xu, H., Dinev, T., Smith, J., & Hart, P. (2011). Information privacy concerns: Linking [4] John, L. K., Acquisti, A., & Loewenstein, G. (2011). Strangers on a plane: Context-de- individual perceptions with institutional privacy assurances. Journal of the Association to collect and process data on any transaction or interaction car- surveys, users indeed wish to have more transparency about data pendent willingness to divulge sensitive information. Journal of Consumer Research, for Information Systems, 12(12), 798-824. ried out on their platform. In consequence, the possession of large collection and usage in order to better assess the benefits and risks 37(5), 858-873. data sets or the ability to collect such data may provide a firm with of data disclosure. Furthermore, one of the guiding policy principles significant data power. In addition, a firm’s competencies with re- of the Digital Single Market strategy of the European Commission spect to data analysis and data processing may also be subsumed is “[f]ostering trust, transparency and ensuring fairness” (EC, 2016). under this concept. In general, it is assumed that such data power However, research does not show a clear effect of transparency on provides a firm with a competitive advantage. For example, the col- user behavior. Although transparency may increase e.g., the effec- lection and processing of large sets of personal data enables online tiveness of personalized advertisements and recommendations, search engines to continually improve their search algorithm and cognitive biases in individual decision-making can also lead to a re- hence the quality of its service, which again may attract more users. duced effectiveness of transparency concepts in practice. Thus, the Moreover, this data enables a search engine to place targeted ad- goal of this thesis is to study the behavioral effects of transparency vertising, which strengthens the firm’s market position in the online in privacy contexts based on an economic laboratory experiment. advertising market. However, whether such competitive advantag-

PROBLEM STATEMENT, RESEARCH QUESTION AND METHODOLOGY

• The goal of the literature survey on the competitive effects of data favor the transition of short-term, data-based competitive advan- collection is to explore the relationship between data power and tages to long-term market power. Based on the findings, the effec- market power and to provide an overview of the factors which tiveness of possible interventions by competition and regulatory

118 | SYMPOSIUM SYMPOSIUM | 119 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

The results so far showed that the amateur and the profession- in real world scenarios. Therefore, for further findings it is expect- al group showed a lower level of cognitive activity during the in- ed that the ICA can be reliably used in an application-oriented real coming-phase compared to the outgoing-phase (in both standard world context, such as badminton. situations). Furthermore, we found that the short time frames of Measuring cognitive load levels professional sports can be tracked validly during active movements during active sport sessions and the use for digitalization PLANNED NEXT STEPS

26 In a next step the reliable measurements of the ICA in an applica- sional) within a learning session. tion-oriented real world context will be tested further and in various To be regarded as a dissertation there is currently still a lot of work other sports-contexts. For example, in a goalkeeper training sce- to be done, especially in the form of more thorough research and Kevin Pfeffel nario in order to develop a training-concept which improves speed publishing results. The aforementioned preliminary study was and accuracy. However, apart from the sports-focus, other appli- merely a first step to prove that the ICA technique can be applied Socio-Informatics and societal aspects of digitalization, Faculty of Computer Science and cations for ICA measurements could be found in human computer in real-world scenarios. Onwards, there are many different applica- Business Information Systems, University of Applied Sciences Würzburg-Schweinfurt interactions. Like a teaching robot who would be able to measure tions to the ICA measurement in various scenarios. As stated, this [email protected] the ICA of a student during automated teaching sessions and would could be either in professional sports training or human computer Prof. Dr. Nicholas H. Müller therefore be able to react to the cognitive efforts by switching the interaction. topic’s difficulty level (for example Beginner, Advanced and Profes- INTRODUCTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION The Index of Cognitive Activation (ICA) (Marshall, 2000, 2007) is still a example, by applying it to driving simulations (Schwalm, Keinath & relatively novel technique of pupil dilation measurement. It doesn’t Zimmer, 2008; Dlugosch, Conti & Bengler, 2013). In a first study went At the moment the project lacks suitable research partners and The expertise of the project exists in the domain of sports. But due use the overall changes in size of the pupil, but instead measures one step further and applied the ICA to examine the cognitive load forms of financing to gather further applicable insights. Therefore, to the existing link to the research field of human computer interac- the frequency of rapid small pupil size variations. Therefore, it is ar- level outside of a laboratory condition by examining badminton play- the biggest question is how to establish suitable research partner- tions, cooperations with experts within the domains of artifical intel- gued that the ICA is more robust to changes in ambient light and ers of the German national team as well as amateurs while being in ships and to identify third-party-funding for a thorough research ligence or machine learning could prove to be useful. eye movements than other techniques. This has been successfully various standard game situations, to prove that the ICA is a reliable project. applied in laboratory conditions (Marshall, Davis & Knust, 2004). For measurement in real world scenarios. REFERENCES PROBLEM STATEMENT AND RESEARCH QUESTION [1] Dlugosch, C., Conti, A. S., & Bengler, K. (2013). Driver Distraction Through Conversation [4] Marshall, S. P., Davis, C., & Knust, S. (2004). The Index of cognitive Activity: estimating Measured With Pupillometry. PROCEEDINGS of the Seventh International Driving cognitive effort from pupil dilation. San Diego, CA: Eyetracking Inc. Technical Report The first problem to solve with this study was to examine if the ICA Within the domain of sports-research, improved training is often- Symposium on Human Factors in Driver Assessment, Training and Vehicle Design, ETI-0401. technique can reliably be used in application-oriented real-world times defined by applying new concepts to standardized situations. 198-204. [5] Schwalm, M., Keinath, A., & Zimmer, H., (2008). Pupillometry as a method for meas- contexts, such as badminton. This is fundamental for further use Therefore, the necessary cognitive functions of applying those [2] Marshall, S. P. (2000). U.S. Patent No. 6,090,051. Washington, DC: U.S. Patent & uring mental workload within a simulated driving task. In D. de Waard, F.O. Flemisch, Trademark Office. B. Lorenz, H. Oberheid, & K.A. Brookhuis (Eds.), Human Factors for assistance and of the ICA technique, for example online measurement of the cog- training concepts is the main focus of interest within my disserta- [3] Marshall, S. P. (2007). Measures of Attention and Cognitive Effort in Tactical Decision automation nitive load level. So, the key research question is how the information. Making. In C. Malcom, J. M. Noyes & Y. Masakowski (Ed.), Decision Making in Complex (pp. 1 – 13). Maastricht, the Netherlands: Shaker Publishing. tion about cognitive load levels is beneficial for scientific research. Environments (pp. 321-332). Ashgate Publishing.

RESEARCH APPROACH AND METHODOLOGY

In order to test the ICA technique, the data of 45 amateur and subsequently second after the racket-contact (outgoing) was used professional players were obtained by having the subjects wear a for analysis. portable Eye-Tracker (SMI ETG2), tracking their pupillary responses The three most important works are from Marshall, Davis and at 60Hz and 120Hz. 38 valid datasets consisting of 28 amateur play- Knust (2004), Schwalm, Keinath and Zimmer (2008) and Dlugosch, ers and 10 professional players were examined. The test subjects Conti and Bengler (2013). These works showed the reliable meas- were asked to play two standard game situations: A serve and a uring of cognitive load levels with the use of the ICA technique in return smash defense. For the analysis, both game situations were different laboratory conditions, for example driving simulations. divided into two different time intervals. The first contact with the Our preliminary tests relied on these findings, but we applied the shuttlecock was used as the starting point and from there a one ICA technique to a real world scenario. second period before the racket made contact (incoming) and the

120 | SYMPOSIUM SYMPOSIUM | 121 SYMPOSIUM ABSTRACTS RELATED WORK

• According to e.g. Samuelson (2000) the absence of intellectual by Swire and Lagos (2013) the right to data portability might have property rights (which may be weakened by a right to data portabil- similar negative effects on firms’ incentives to innovate and create ity) might harm society due to reduced incentives to invest in data databases. creation. Zhu et al. (2008) investigate the incentives of a database’s • However, the right to data portability in conjunction with Personal creator if another firm can reuse the underlying data. They find that Data Markets does not only entail direct effects on market partic- a low level of differentiation between the original database and the ipants, but also induce additional strategic implications leading to database of a free-riding competitor might discourage the former counterintuitive findings. The contribution of this study is to iden- Empowering Users in to create the database in the first place. In this context, compensa- tify these effects and to provide a framework for regulatory inter- 27 the Data Economy tion of the data creator might help to overcome this issue. As noted ventions. PRELIMINARY RESULTS

Bastian Haberer • A Personal Data Market that pays users for access to their data orig- additional competition and ii) in order to induce the Personal Data inally created by a platform and monetizes these data by providing Market to subsidize usage through a positive price for consumers’ University of Passau, School of Business, Economics and Information Systems, Research Group Data Policies similar data services exerts competitive pressure on the platform data. [email protected] · http://www.datapolicies.net which might reduce the latter’s profits. • Even though content providers lose exclusive access to data creat- • If a Personal Data Market is able to generate a relatively high rev- ed at their website, they might benefit from Personal Data Markets, Dr. Daniel Schnurr · Prof. Dr. Jan Krämer enue from data services, it might have an incentive to boost data if a consumer’s additional monetary incentive to create data out- creation on a platform by remunerating consumers for the creation weighs negative implications from increased competition. of data. In consequence, consumers may increase their usage. • Consumers might actually be worse off when remuneration does • Platforms might have an incentive to reduce quality for two rea- not outweigh reduced service quality in the presence of the Person- INTRODUCTION sons: i) because of reduced returns on quality investment due to al Data Market.

In the online ecosystem, platforms often engage in a business model should be remunerated for data creation akin to labor (Arrieta Ibarra in which they serve two distinct customer groups: consumers and et al. 2017). Moreover, e.g. via the General Data Protection Regula- users of data services. Whereas consumers are typically natural per- tion (GDPR) and the newly introduced right to data portability, the EU sons, data service users are companies interested in e.g. placing tar- has undertaken first steps towards a more user-centric data econ- PLANNED NEXT STEPS geted advertisements or generating insights through data analytics. omy. Many online platforms offer their services to consumers free of This new regulatory framework has the potential to seriously affect • Within the next months, extended versions of the model will be • For future work, I consider to do research on the incentives of users charge. However, by using these services, consumers create data. the online economy as a whole. By now, however, there has been analysed to test the robustness of the main model and gain more to create and sell high-quality data in the context of different stated These data are then used by the platform to offer data services to little research on the economic effects of such kind of regulation on nuanced insights. purposes of data collection. the data service users. consumers and companies in online markets. In this context, scholars have raised the question, whether users OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

PROBLEM STATEMENT AND RESEARCH QUESTION Collaborations with technical experts would be helpful to better un- learning outcomes. Besides that cooperation with legal experts derstand the effects of data quantity and data quality on machine might be fruitful to turn insights into regulatory practice. • Current policy proposals and new regulatory rules that aim at em- - Under which conditions do users benefit from Personal Data Mar- powering users in the data economy require research on the actual kets? economic effects of user empowerment on market outcomes. This - Can platforms profit from this kind of user empowerment? REFERENCES thesis seeks to eludicate these issues. • By addressing these questions, the study provides insights into con- • More specific, a game-theoretic model is devised to investigate the sumers’ incentives to provide information and contributes to the [1] Arrieta Ibarra, I., Goff, L., Jiménez Hernández, D., Lanier, J., and Weyl, E. G. 2018. [4] Zhu, H., Madnick, S. E., and Siegel, M. D. 2008. "An Economic Analysis of Policies for the effects of market places where consumers may sell personal data understanding of the basic economic effects of a user-centric data "Should We Treat Data as Labor? Moving Beyond 'Free'," American Economic Associa- Protection and Reuse of Noncopyrightable Database Contents," Journal of Manage- on their own – so called Personal Data Markets. In particular, the economy. tion Papers & Proceedings (108) pp. 38-42. ment Information Systems (25:1), pp. 199-232. [2] Samuelson, P. 2000. "Privacy as Intellectual Property?," Stanford Law Review (52:5), study examines the following research questions: • Besides investigating potential regulatory pitfalls, this study also of- pp. 1125-1173. - How are service quality and prices in digital markets affected by fers managerial implications on how companies in digital markets [3] Swire, P., and Lagos, Y. 2013. "Why the Right to Data Portability Likely Reduces the emergence of user-centric Personal Data Markets? should deal with more user empowerment. Consumer Welfare: Antitrust and Privacy Critique," Maryland Law Review (72:2), pp. 335-380.

RESEARCH APPROACH AND METHODOLOGY

• So far, the research is built upon a game-theoretic model. Using a a service to these consumers and, on the other hand, offer data game-theoretic model has the advantage of focusing on the stra- services to iii) data service users. In the model, consumers decide tegic effects key to answering the research questions. Moreover, about their level of usage of the platform’s service and, thereby, im- it might help to test new regulatory proposals, where no data is plicitly on the amount of data that is created at the platform. Firms available or if those proposals are not yet implemented in practice. can invest in service quality, potentially charge users a price and Thus, this research method is suited to identify the main effects process users’ data for the provision of data services. of regulatory reforms in the first place and to clarify the involved • This set-up allows to identify effects of a usercentric Personal Data economic trade-offs. Market on service quality, data service quality, prices, firms’ profits • In particular, the set-up of the game-theoretic model in the study and consumer welfare. consists of the three kinds of actors that are characteristic for digital markets: i) consumers, ii) platforms that, on the one hand, provide

122 | SYMPOSIUM SYMPOSIUM | 123 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

This work’s research approach is primarily categorized as systems data might add that users dislike the fastest way and prefer a slower research. The whole work is separable into cycles which iteratively version. build on one another. Once a new prototype or prototype state is developed based on the findings and insights of the previous one, In terms of related work, Isenberg et al.‘s proposed Buffer Framework this iteration is put to the test with users (lab and public administra- (2006) and its mobile spatial tools extension (Isenberg et al., 2007) tion facilities). These user studies can be conducted in various ways primarily inspire my work. I expand on this in three ways. First, I and are depending on the goal of the implementation. For example, achieve more flexibility for the users by removing the constraint that Sketchable Workspaces if a whole prototype is tested it can be quantified how long specific a buffer is either placed by the system or stuck to a tool. Users are task completion times for workflow modeling are. Also actual work- empowered to arbitrarily sketch regions themselves and where they and Workflows flow completion times can be logged with the system, once broken want to. The second extension is that actions are not limited to only down to recurring task primitives such as reviewing a pile of files. changing the visual representation of objects such as file icons on 28 Accumulating these data with quantitative interview data tops off the the screen, the objects‘ data are also manipulated. The third exten- feedback for this cycle. If a single feature sion is that objects are not only just digital artifacts. The idea that in- (interaction technique) is subject to evaluation then its suitable to teractive regions also may affect windows is inspired by “Everything Jürgen Hahn provide multiple different versions measuring with which one users Is a Window“ (Wimmer and Hennecke, 2010). are the fastest, in order to maximize efficiency. However, qualitative University of Regensburg, Physical-Digital Affordances Group [email protected] Dr. Raphael Wimmer PRELIMINARY RESULTS

The early proof of concept implementation, as seen in Figure 1, is conducting user studies. Quantitative measurements like evaluat- almost completed and ready for user tests. Hence, there are no ing task completion times are yet to be done. However, I expect INTRODUCTION concrete findings from an empirical perspective, yet informal try- that novice users will start rather slowly, until they grasp the sys- out sessions with colleagues yielded very positive reactions and tem’s possibilities and their customization opportunities. The goal of my research is to develop, implement and evaluate can be dragged over such a region and its contents are shown in a great feedback. I expect to gain similar qualitative results when Sketchable Workspaces and Workflows, a novel concept for visual preview or is deleted or tagged or sent via email. The effects may end-user programming. It aims to empower end-users to intuitively apply to a multitude of objects – including digital artifacts and their customize their workspaces and workflows according to their prefer- graphical representation (“file icons”), windows of the operating sys- ences, in order to enhance their performance and satisfaction. Users tem, pointers (mouse, user’s fingers, pens), and physical objects such can sketch interactive regions on their computer desktop with their as paper documents, smart phones and other devices, or tangible PLANNED NEXT STEPS fingers, a pen, or mouse. The primary focuses areinteractive surfaces widgets. Users can build complex such as digitally augmented physical desks or multi-touch tabletops. setups and tools and tool-chains for their everyday office work tasks The user study for the current prototype will be conducted next. Af- evaluation stages will start. In order to put my work into a disserta- These usergenerated interactive regions apply specific effects to by combining the necessary regions and objects. Figure 1 shows ter the study, I migrate the system to an actual desk which will be tion, a lot of focusing and streamlining appears to be required once any object touching them, thereby modifying the properties of their such a user-generated workflow, created with the current prototype augmented with over-head projection and change certain aspects the broad exploration phase is concluded. visual representation or the underlying data. For example, a file icon implementation on a multi-touch table (Samsung SUR40). based on user feedback. Then the cycle based development and

PROBLEM STATEMENT AND RESEARCH QUESTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

The status quo in public administration office work shows a distinct as and worth of further abstraction. Hence, a contribution of this Some aspects of the implementation of the over-head tracking are gap between the digital and physical worlds in regard of workflows work is to provide a concept and tool for researchers and practi- still unclear. Also, the user studies are not fully planned out yet and or processes. Typically, this work involves a lot of physical records tioners which also inspires new ideas. This work also contributes a subject to discussion. or paper in general, but also digital files which sometimes can car- notation for workflows, enabling users to formalize their processes ry the same information. The physical part of the workflow has to visually and iteratively improve them based on their sketches. Ad- be brought into the digital space. However, users tend to lose con- ditionally, this work contributes proof of concept implementations REFERENCES trol of their workflows in digital systems, because these systems which may be deployed in actual office environments and software are traditionally very limited in terms of customization. Users are packages on which others may build new applications. [1] Isenberg, Tobias, André Miede, and Sheelagh Carpendale. 2006. “A Buffer Framework [3] Wimmer, Raphael, and Fabian Hennecke. 2010. “Everything Is a Window: Utilizing the trapped in inflexible user interfaces which are partially incompati- for Supporting Responsive Interaction in Information Visualization Interfaces.” In Window Manager for Multi-Touch Interaction.” In Workshop ”Engineering Patterns for ble to or difficult to integrate with physical workflows. Yet, this kind Creating, Connecting and Collaborating Through Computing, 2006. C5’06. the Fourth Multi-Touch Interfaces” in Conjunction with ACM EICS 2010 International Conference on, 262–69. IEEE. of office work always requires physical and digital documents. [2] Isenberg, Tobias, Simon Nix, Martin Schwarz, André Miede, Stacey D. Scott, and Sheelagh Carpendale. 2007. “Mobile Spatial Tools for Fluid Interaction.” 2007-872-24. Solving this problem means that individual users regain lost flex- Canada: Department of Computer Science, University of Calgary. Doi:1880/45785. ibility and also potentially efficiency if certain user interfaces are so generic that the users‘ day-to-day business is impeded by that. From an economic standpoint, complex processes require skilled workers and constant improvement, yet e.g. public administration employs inflexible systems which are hard to change, causing loss of efficiency and flexibility.

My primary research question is how to build a software system that is generic enough to allow users to intuitively design custom workflows for their workspaces. Which properties should it have? Which Figure 1: Prototype Implementation running on a multi-touch table. Here, two file icons are trans- interaction techniques work and which ones do not? Furthermore, it ported across the screen by a looping digital conveyor belt running through three interactive regions. One of the files collides with the red coloured region, so its content is shown as aseamless has to be determined how well the system scales to complex work- preview, while the other one (digital twin of the sheet of paper) previously crossed the Send-via- flows and multiple users, and which elements are actually good ide- email region and therefore is currently being sent.

124 | SYMPOSIUM SYMPOSIUM | 125 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

The Integrated Behavioral Model has already been studied for its both projects were applied and evaluated in a university at the be- applicability in the information security context [15, 16]. A propos- ginning of the year and allowed a first analysis of the ISA of university al of a complementary process that targets ISA of SME’s employees members. was developed based on the ISIS12 framework [4]. The findings from Sensitizing employees of small and medium-sized companies individually PLANNED NEXT STEPS for information security The next step will be to finalize the summary of ISA basics in the form were able to increase ISA is evaluated. The findings from the univer- 29 of a working report later this year. In the abovementioned university sity project will be used to revise the approach and apply it to the project, ISA-techniques are to be identified that enable an increase partner companies next year. of ISA, thereby considering the results of the ISA-analysis of the uni- Andreas E. Schütz versity members. Subsequently, the extent to which the techniques University of Applied Sciences Würzburg-Schweinfurt, Faculty of Computer Science and Business Information Systems in cooperation with University of Erlangen-Nuremberg OPEN ISSUES AND POSSIBILITIES FOR COOPERATION [email protected] · www.fhws.de Prof. Dr. Kristin Weber · Prof. Dr. Felix Freiling · Prof. Dr. Nicholas Müller Feedback and testimonials from the people in charge of Informa- on the efficiency of individual measures in advance. In addition, we tion Security at companies, especially SME, that have already suc- are still looking for ways to fund research. cessfully carried out ISA campaigns would be helpful for the further INTRODUCTION course of the project. This would help to gain practical experience

Small and medium-sized enterprises (SME), that constitute more with techniques like phishing, malware, and social engineering [5, p. than 99 percent of German companies [1, p. 2], are affected by the 11]. For attackers, finding out a password from an employee is by far REFERENCES increasing digitization. Digitized business processes challenge com- easier than "hacking" it with a compute-intensive brute-force attack. panies, e.g., there is a growing demand for information security [2]. In addition, SMEs are increasingly affected by cyber-attacks [3]. With [1] BMWi: Aktionsprogramm Zukunft Mittelstand. Neuauflage 2016. 2016. [11] Jarvinen, P. H., Research Questions Guiding Selection of an Appropriate Research Employees play a key role in information security. The behavior of regard to the lower budget of SMEs, it is necessary to offer them [2] BMWi: Monitoring-Report Wirtschaft DIGITAL 2016. 2016. Method, Ecis 2000, S. 124-131. 2000. employees at work and in their private life strongly affects the con- cost-effective support to sensitize their employees. [3] PwC: Im Visier der Cyber-Gangster – So gefährdet ist die Informationssicherheit im [12] Hevner, A. R., March, S. T., Park, J., Ram, S., Design Science in Information Systems deutschen Mittelstand. 2017. Research, in: MIS Quarterly, 28, 2004, Nr. 1, S. 75-105. 2004. fidentiality, integrity and availability of sensitive company informa- [4] Weber, K.; Schütz, A.: ISIS12-Hack: Mitarbeiter sensibilisieren statt informieren. [13] Montaño, D.; Kasprzyk, D.: Theory of Reasoned Action, Theory of Planned Behavior, tion [3, S. 16; 6]. Attackers deliberately exploit this "human factor" In: Multikonferenz Wirtschaftsinformatik 2018, 6.-9.3.2018, Lüneburg. 2018. and the Integrated Behavior Model. In: K. Glanz, B. Rimer & K. Viswanath: Health [5] ISACA (2017): State of Cybersecurity – Part 2: Current Trends in the Threat Landscape. Behavior and Health Education. Theory, Research and Practice. 4th Edition, John Wiley 2017. & Sons. Hoboken. S. 67-96. 2008. [6] Albayrak, C.; Gadatsch, A.: Sind kleinere und mittlere Unternehmen (KMU) bereits [14] Checkland, P., Holwell, S.: Action Research: Its Nature and Validity. Syst. Pract. auf die Digitale Transformation vorbereitet? In: Multikonferenz Wirtschaftsinformatik Action Res. 11. 1998. PROBLEM STATEMENT AND RESEARCH QUESTION 2018, 6.-9.3.2018, Lüneburg. 2018. [15] Schütz, A.: Sensibilisierung der Mitarbeiter von kleinen und mittleren Unternehmen [7] Helisch, M.; Pokoyski, D.: Security Awareness. Neue Wege zur erfolgreichen für das Thema Informationssicherheit. Masterarbeit, FHWS – Hochschule für ange- Mitarbeiter-Sensibilisierung. Vieweg+Teubner. Wiesbaden. 2009. wandte Wissenschaften Würzburg- Schweinfurt. 2017. To make employees aware of their important role in information se- considers the specific needs of SMEs and supports individuality at the [8] Schroeder, J.: Advanced Persistent Training – Take Your Security Awareness Program [16] Schütz, A.; Weber, K.: Security Awareness: Nicht nur schulen – überzeugen Sie! curity, they need to be sensitized to behave securely [7; 4]. We call this same time. to the Next Level. Apress. 2017. In: Schartner, P. (Hrsg.): D-ACH Security 2017. syssec. München. 2017 behavior information security compliant behavior. The sensitization of [9] Hassel, S.: Fallstudien zu Security Awareness und dem Management der Informations- sicherheit in kleinen und mittleren Unternehmen. Masterarbeit, FHWS – Hochschule employees (also Information Security Awareness, ISA) results from the These findings raise the following main research question: How can für angewandte Wissenschaften Würzburg-Schweinfurt. 2017. interplay of cognition (understanding of the problem and the knowl- SMEs be supported increasing the ISA of their workforce? In addition, [10] Österle, H. et al.: Memorandum zur gestaltungsorientierten Wirtschaftsinformatik. In: Zeitschrift für betriebswirtschaftliche Forschung, 6, 2010, Nr. 62, S. 664-672. 2010. edge to solve it), intention to act (will of the employee to behave in ac- the following secondary research questions arise: To what extent cordance with information security) and the organization (Employees can technical measures minimize the involvement of the workforce can behave in compliance with information security in their respec- through information security compliant behavior? What is the cost/ tive environment) [7]. Experience shows that ISA-raising concepts are benefit ratio for ISA measures? What possible negative consequences successful if they are specifically tailored to the needs of the compa- can be caused by increasing ISA? ny and its employees [8]. Such concepts can only be developed and implemented with high financial effort by the companies. However, The aim of the work is to develop a methodology that enables SMEs SMEs often have limited financial resources for information security to sensitize their employees for information security in a targeted and [9, S. 4]. Currently, there is no methodological approach to ISA that sustainable way involving little financial, personal and time effort.

RESEARCH APPROACH AND METHODOLOGY

The doctoral project is assigned to the design-oriented information in health psychology and is interpreted in the context of information systems research according to [10] and follows the artifact-creating security in the project. Subsequently, possibilities for measuring the research paradigm "Design Science" [11, S. 124f]. Design Science re- degree of sensitization of a company are examined and how ISA-in- search produces useful solutions that are developed through the decreasing techniques can purposefully influence the behavior of em- sign and evaluation of artifacts, like constructs, models, methods or ployees. Most of the research will take place directly at our 14 partner systems. Design Science thus contributes to both fulfilling concrete companies. That is, we follow the action research approach, where entrepreneurial requirements and increasing the scientific knowledge researchers and the cooperating institution jointly develop a problem [12, p. 81]. The iterative process ensures rigor of research results. solution [14]. We will also analyse the economic viability of ISA-techniques. Subsequently, the findings are arranged and summarized in As a fundamental theory of human behavior, the Integrated Behav- a methodology. ioral Model (IBM) [13, p. 77] is used in this work. The IBM originates

126 | SYMPOSIUM SYMPOSIUM | 127 ABSTRACTS bution gaps (e.g., Danaher, 2017) and to responsibility shifting (Bart- & Gächter, 2000), and anti-social punishment (Herrmann et al., 2008). SYMPOSIUM ling & Fischbacher, 2011), cooperation-promoting institutions (Fehr

PRELIMINARY RESULTS

We are currently developing the comprehensive research frame- unethical outcomes can be effectively shifted to other persons; c) an- Experimental Studies on Filling work for our laboratory experiments. Based on this framework, we tisocial punishment is widespread. From these results we conclude will design new experimental treatments. Previous works (cited ab- that people will be willing to punish scapegoats for filling AI-induced AI-Induced Retribution Gaps through ove) have shown that: a) The retribution gap arises from a mismatch retribution gaps. between the human desire for retribution and the absence of appro- 30 Scapegoat Punishment priate subjects of retributive blame, b) responsibility attribution for Gari Walkowitz PLANNED NEXT STEPS Technical University of Munich, TUM School of Governance, Digital Ethics Group We will finalize the theoretical analysis and the design phase for our sessions and present the results for discussion. [email protected] · www.gov.tum.de laboratory experiments. We will then run the first pilot laboratory Dr. Matthias Uhl · Prof. Dr. Christoph Luetge OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

INTRODUCTION We want to discuss the complexities of our research designs with the scholars who are experts in AI-induced wrongdoing. ZD.B experts and audience. We would like to cooperate with legal Legal implications of automation cial responses are widely underexplored. The Bertelsmann-Stiftung Many experts identify severe liability gaps due to automation. For (2018) identifies a strong aversion against the use of algorithms. Part example, Shawn Bayern was the first to show how current company of the aversion against algorithms could stem from the “retribution REFERENCES law in US can be used to establish legal person, i.e., a company that gap” (Danaher, 2017). is solely controlled by artificial intelligence (AI). In 2017, European [1] Bartling. Björn & Fischbacher. Urs. Shifting the blame: On delegation and responsibili- the social sciences. Science, 326(5952), 535-538. 2009. Parliament urged European Commission to propose “electronic per- The retribution gap ty. The Review of Economic Studies, 79(1), 67-87. 2011. [7] Fehr. Ernst & Gächter. Simon. Cooperation and punishment in public goods experi- sonality” for sophisticated autonomous robots. Further, national law People are innate retributivists (Carlsmith & Darley, 2008). With in- [2] Bertelsmann Stiftung. Was Deutschland über Algorithmen weiß und denkt. ments. American Economic Review, 90(4), 980-994. 2000. must require AIs to indicate a responsible natural person, e.g., on a creased automation, more social harm will be induced by artificial https://www.bertelsmann-stiftung.de/de/publikationen/publikation/did/was-deutsch- [8] Harrison. Glenn. W. & List. John. A. Field experiments. Journal of Economic Literature, land-ueber-algorithmen-weiss-und-denkt/. 2018 42(4), 1009-1055. 2004. letterhead or in a register (Burri, 2017). agents. However, automated agents are unlikely to meet conditions [3] Burri. Thomas. Free Movement of Algorithms: Artificially Intelligent Persons Conquer [9] Herrmann. Benedikt., Thöni. Christian & Gächter. Simon. Antisocial punishment across of retributive blame. Therefore, there will be a mismatch between the European Union's Internal Market, Discussion Paper. 2017. societies. Science, 319(5868), 1362-1367, 2008. [4] Carlsmith. Kevin. M. & Darley. John. M. Psychological aspects of retributive justice. Ethical implications of automation modern reality and: a) the legal necessity of responsibility ascription, Advances in Experimental Social Psychology, 40, 193-236. 2008. The ethical implications of harm being caused by algorithms and so- b) people‘s psychological desire for retribution. [5] Danaher. John. Robots, law and the retribution gap. Ethics and Information Techno- logy, 18(4), 299-309. 2016. [6] Falk. Armin & Heckman. James. J. Lab experiments are a major source of knowledge in PROBLEM STATEMENT AND RESEARCH QUESTION

The aim of our study ner)? Do wrongdoers actively shift responsibility to scapegoats? Who We theoretically and experimentally study the social consequences agrees to be a scapegoat? of AI-induced retributions gaps. We investigate whether AI liability can be effectively shifted to natural persons (scapegoats) who are Academic and practical contribution held responsible, i.e., we separate legal and ethical wrongdoing from The emerging unfilled retribution gaps might cause severe social the recipients of social punishment. problems − legally and psychologically. Our study is very important for an understanding and maintaining of cooperative social sys- Main research questions tems under digitization. It contributes to the literature on cooperati- In our project, we want to answer the following research questions: on-promoting institutions and anti-social punishment (Fehr & Gäch- Under which circumstances are people willing to punish scapego- ter, 2000, Herrmann et al., 2008). It also sheds light on the causes ats? Do punishers experience relief or satisfaction when sanctioning of algorithm aversion and informs the development of institutions scapegoats? How does the existence of a scapegoat influence the to overcome social resistance against AI-driven technology and for propensity of an actor to do wrong (e.g., an AI’s programmer or ow- stabilizing societal cooperation in AI-enhanced environments.

RESEARCH APPROACH AND METHODOLOGY

Behavioral experiments to study behavioral mechanisms Complementary use of other methods We combine state-of-the-art behavioral theory and empirical met- Self-report measures, e.g., surveys and interviews, are complemen- hods to study scapegoat punishment by analyzing and directly ob- tarily used. Self-report measures are, however, potentially biased, serving people’s behavior. Controlled laboratory studies enable a e.g., by socially desirable responding, and behavioral mechanisms direct investigation of the mechanisms and influencing factors which cannot unambiguously be separated due to confounding factors. drive scapegoat punishment (Falk & Heckman, 2009). Field experiments allow studying scapegoat punishment in real-life environ- Related literature ments (Harrison & List, 2004). Our work directly connects to the recent works on AI-induced retri-

128 | SYMPOSIUM SYMPOSIUM | 129 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

In the past months, we created ideas towards supporting interac- movements). We expect to identify variations and uniqueness in tion in everyday life using feedforward concepts [11] (i.e., guiding behavioural patterns for several interaction metaphors, which are and informing users about possible actions and results [5]) and feasible to apply for (behavioural) biometric authentication in VR. supporting older adults concerning privacy and security settings on In addition, VR offers great opportunities as a study methodology mobile devices by utilising related behaviour of trusted individuals itself, since we can collect data in a controlled setup without expos- (e.g., children and grandchildren) [9]. As a “special life” use case, ing participants to real, potentially dangerous situations (e.g., road we investigate user behaviour in virtual reality (VR). In a pilot study, traffic, heights), being able to observe their behaviour and context Supporting Users in Everyday we observed that users indeed tend to behave differently (e.g., from a variety of viewpoints (e.g., 3D tracking data, movement, eye 31 Life Using Behavioural Biometrics one vs both-handed use of controllers, different amounts of head gaze, manual observations). Sarah Prange PLANNED NEXT STEPS As a next step, we plan to conduct a field study in users’ (smart) (potentially unique) behaviour in such situations using various track- Munich University of Applied Sciences, Department of Computer Science homes. The aim is to find out which devices play central roles in using technologies. Moreover, I will soon join Prof. Nigel Davies’ group [email protected] ers’ daily behaviours and routines and which of them are feasible to at the Data Science Institute at Lancaster University for a three- PhD at LMU Munich, LFE Media Informatics · https://biometrics.medien.ifi.lmu.de / use for recognizing unique, identifiable behaviour patterns. In the month research internship. I will work on two projects among the EU Prof. Dr. Florian Alt · Dr. Emanuel von Zezschwitz context of a bachelor thesis, we are currently exploring interaction funded PETRAS1 project, in particular on privacy and trust in the era behaviour in problem solving tasks, inspired by tasks in daily life (e.g., of the Internet of Things, covering applications of usable security in cooking). We simulate common problem types with LEGO assembly everyday life (e.g., public displays). INTRODUCTION tasks, which represent daily situations (e.g., users do not know how 1 https://www.petrashub.org/ to blanch vegetables or miss an ingredient). We investigate users’

Users frequently need to apply various kinds of authentication unobtrusive authentication. While many research projects on behav- mechanisms to gain access to something in their daily life, such as ioural biometrics exist (e.g. on gait [4] or keystroke dynamics [2]), we buildings, cars, devices or data. Authentication is present in everyday rarely find behavioural patterns measured and applied as (exclusive) OPEN ISSUES AND POSSIBILITIES FOR COOPERATION life and is mostly burdensome for users [10]. An important question authenticators in common daily use so far. Behavioural patterns in human-computer-interaction (HCI) is how to bridge the gap be- could not only facilitate authentication, but also offer great potential Firstly, understanding the basics of human behaviour is essential volve collaborations with the new Cyber Defence Research Centre tween usability and security. Behavioural biometrics identify users for applications beyond security purposes. During my PhD, I rethink for my research. Within this context, I did not stick to HCI litera- at Bundeswehr University, Munich or other experts within the ZD.B by “something they do” [12] (i.e., by behavioural aspects rather than authentication and biometrics for applications in everyday life. ture, but conducted literature review in a broader scope. Hence, network. For implementing concrete biometric systems or analys- physiological features) and provide great opportunities for implicit, research in behavioural biometrics not only offers great potential ing behavioural data, we might in the future need more expertise for diverse applications in various fields (e.g. health [1, 7]), but also (e.g. in cryptography, machine learning) which again offers possibil- for cooperation among those disciplines. In particular, we will join ities for collaborations with experts (from research as well as from PROBLEM STATEMENT AND RESEARCH QUESTION a joint project between the LMU Department of Psychology and the industry). Finally, research in the areas of the Internet of things as Media Informatics Chair soon, improving mobile experience sam- well as smart home applications offers great opportunities for a pling for psychological studies, collecting behavioural data with a research or industry driven collaboration. I will start working and My main goal is to understand behavioural patterns in various use “behaviours” are not feasible for identifying users among the whole smartphone application. Secondly, implementing secure systems collaborating in this direction during my research internship at Lan- cases, particularly in casual, everyday life. Be it “natural” behaviour, target group (i.e., not applicable for critical security cases), they processing sensitive user data to support their everyday life re- caster University. interaction with people and devices, or reactions to certain stimuli might be sufficient for casual, everyday life use cases involving less quires a deeper knowledge of actual cyber security. This might in- and events, it is of great interest to find those patterns which are severe forms of authentication (e.g., to verify which roommate is feasible for biometric applications. While biometric features have approaching the coffee machine). I further investigate and analyse to fulfil several characteristics for authentication use cases [8], it existing problems with authentication-related interactions, espe- might be of interest to distinguish users via “soft biometric” features cially those which could be facilitated by the means of (behavioural) REFERENCES [3] (e.g., height, age, gender) for everyday life use cases. Regard- biometrics. Consequently, I explore behavioural biometric applica- ing behavioural features, we could classify people using common tions for support in everyday life. This not only covers the design [1] Johan Andrén and Peter Funk. A Case-Based Approach Using Behavioural Biometrics [7] Marcos Faundez-Zanuy, Amir Hussain, Jiri Mekyska, and others. Biometric applications to Determine a User’s Stress Level. In ICCBR Workshops. 9–17. 2005. related to human beings: there is life beyond security. Cognitive Computation 5, 1 behaviours and routines (e.g., morning grumpiness vs happiness and implementation of such applications, but also evaluating these [2] Salil P Banerjee and Damon L Woodard. Biometric authentication and identification (2013), 136–151. 2013. might influence their morning routine and behaviour). While such regarding usability and security. using keystroke dynamics: A survey. Journal of Pattern Recognition Research 7, [8] Anil K. Jain, Karthik Nandakumar, and Arun Ross. 50 years of biometric research: 116–139. 2012. Accomplishments, challenges, and opportunities. Pattern Recognition Letters 79, [3] Antitza Dantcheva, Petros Elia, and Arun Ross. What Else Does Your Biometric Data 80–105. 2016. Reveal? A Survey on Soft Biometrics. IEEE Transactions on Information Forensics and [9] Lukas Mecke, Sarah Prange, Daniel Buschek, Mohamed Khamis, Mariam Hassib, Flori- Security 11, 3, 441–467. 2016. an Alt. “Outsourcing” Security: Supporting People to Support Older Adults. MobileHCI RESEARCH APPROACH AND METHODOLOGY [4] M. O. Derawi, C. Nickel, P. Bours, and C. Busch. Unobtrusive User-Authentication on ’18 Workshop on Mobile Privacy and Security for an Aging Population. 2018. Mobile Phones Using Biometric Gait Recognition. In 2010 Sixth International Confer- [10] Shrirang Mare, Mary Baker, and Jeremy Gummeson. A study of authentication in daily ence on Intelligent Information Hiding and Multimedia Signal Processing. 306–311. life. In Proceedings of SOUPS, Vol. 6. 2016. In our research, we evaluate both, the usability as well as the security biometric technologies. To find behavioural patterns in everyday life, 2010. [11] Sarah Prange, Daniel Buschek, Florian Alt. Everyday Life Feedforward. CHI ’18 Work- of new concepts. For usability aspects, we value user centred design we especially plan to conduct field studies since this brings us closest [5] Tom Djajadiningrat, Kees Overbeeke, and Stephan Wensveen. 2002. But How, Donald, shop on Rethinking Interaction. 2018. processes and usability evaluations. For security aspects, we con- to participants’ life. In addition, experience-sampling methods will Tell Us How?: On the Creation of Meaning in Interaction Design Through Feedforward [12] Roman V. Yampolskiy and Venu Govindaraju. Taxonomy of Behavioural Biometrics. and Inherent Feedback. In Proceedings of DIS ’02. In Behavioral Biometrics for Human Identification. IGI Global, 1–43. 2010. duct attack studies evaluating concepts against certain threat mod- help us to collect in-situ user data via qualitative feedback or sen- [6] Malin Eiband, Mohamed Khamis, Emanuel von Zezschwitz, Heinrich Hussmann, and els (e.g., shoulder surfing [6]). We often evaluate early prototypes sor-based context data. Florian Alt. Understanding Shoulder Surfing in the Wild: Stories from Users and Observers. In Proceedings of CHI ’17. in explorative studies to investigate how people will perceive new

130 | SYMPOSIUM SYMPOSIUM | 131 ABSTRACTS by the brain to the muscles to perform a particular movement? While the model can also provide important insights into the field SYMPOSIUM How these signals differ between the people and between the simi- of motor control, the current focus is to make it practical rather then lar movements? How the signals are adapted to the particular state theoretically exact. of the musculoskeletal system, in particular to the posture, the cur- The method can accelerate and improve research in post-desktop rent movement, the state of the muscles and fatigue? How the ex- interfaces, exergames, and it might contribute to the fields of bi- citation signals will fit into the forward biomechanical simulation? omechanics, motor control and rehabilitation. It might be incorporated into the currently existing biomechanics and ergonomics Control-theoretical models for The expected contribution is a model/algorithms/controller to gen- software, as well as it can be implemented in a stand-alone soft- erate the neural excitation signals which while fed into forward bi- ware for movement design. biomechanics-consistent control omechanical simulation would produce a human-like movement. of arm movements in HCI 32 RESEARCH APPROACH AND METHODOLOGY In our research we follow the scientific method and use quantita- mathematical derivation and proofs can contribute to a small extent, tive experimental data collection, use train data for model fitting and but due to the complexity the model needs to be data-driven. Dr. Myroslav Bachynskyi simulation, and evaluate the outputs against the test data. We use Bayreuth University, Faculty of Mathematics, Physics and Computer Science existing datasets as well as collect our own motion capture, force My research is based on biomechanical simulation and neuromus- and EMG data. While we use control theoretical models for gener- cular control research from Stanford, the University of Tennessee, [email protected] ating the neural excitation signals, alternatively we can try to train a and the University of Washington. The closest works in HCI were Prof. Dr. Jörg Müller deep neural network to generate the corresponding signals or even published by Prof. Dr. Müller, me and our collaborators. Few oth- to directly generate biomechanical variables of other levels. Howev- er works within HCI inversely simulate “primitive” biomechanical er, such model, even if successfully providing high-accuracy outputs, models without muscles or control. No other works have generat- cannot be considered other as a black box preventing from generated human-like movement for the complete musculoskeletal system INTRODUCTION ing insights of the internal signal generation. On the other side, pure neither in HCI nor in biomechanics.

This abstract describes a new approach to analysis and evaluation Classical ergonomics methods like EMG, or goniometry do not work of physical ergonomics of mid-air movements. It will be able to pre- for such evaluation due to huge design and movement spaces and PRELIMINARY RESULTS dict human-like movements and all internal states of musculoskel- complexity of their application. Additionally, they provide data only etal system for a particular aimed movement. Based on the initial for small set of muscles or joints within the human body, which makes In our previous works we have identified repeating patterns with- we have integrated biomechanical simulation framework OpenSim posture and the target to be reached by the end-effector our con- it necessary to know the regions under risk in advance. Previously in muscle activation signals within various whole arm aimed move- with Simulink for a small musculoskeletal model. We plan to tune trol-theoretical model will be able to predict the neural excitations, motion capture-based biomechanical simulation was proposed as a ments; we have generated human-like end-effector trajectories, Simulink model to generate the neural excitations, while the forward which as inputs run the forward dynamics musculoskeletal simu- method for efficient ergonomics evaluation of post-desktop interfac- velocities and accelerations using control theoretical models; and dynamics will be simulated by OpenSim. lation to generate the necessary musculoskeletal movement. The es [3]. The optical motion capture accurately tracks kinematics of hu- simulation generates all intermediary states of the musculoskeletal man movement in 3D space and together with external forces data system, including activations of muscles and corresponding muscle provides complete description of mechanical output of human body. forces, joint moments and angles, end-effector position velocity and However, to interpret this data in context of human body, namely acceleration for the whole movement trajectory. All these variables as the variables describing the within-body mechanical loads we PLANNED NEXT STEPS can be used in further analysis of physical ergonomics of the move- need to apply biomechanical simulation. This simulation uses accu- ment. rate model of musculoskeletal system of human as a prior to fit the Tune and train the Simulink model on an existing dataset of aimed and validate the outputs. Collect movement data for complex move- data. Then it can provide information concerning the other variables movements. Record EMG data of aimed movements and validate ments and derive the decomposition of them into the simple aimed Full-body interfaces become a lot of attention in the HCI field, in within the model, namely joint angles and moments, muscle forc- the generated by Simulink signals against the data. Collect motion movements. Find inverse to the decomposition and apply it to the particular in the recent years with the advancements in VR headsets es and activations, or even neural excitations [4]. The biomechani- capture data, run the Simulink model together with biomechanics model. and technology. The VR technology offers many benefits for various cal simulation produces these results in multiple consecutive steps, application scenarios from medical treatment [1] to games or even namely inverse kinematics, inverse dynamics, static optimization reduction of implicit racial bias [2]. However, the large problem of and computed muscle control. Most of these steps involve complex those interfaces is the complexity of appropriate physical ergonom- computations and take long time to produce results. This processing OPEN ISSUES AND POSSIBILITIES FOR COOPERATION ics evaluation. Without well-analyzed physical ergonomics the VR can pipeline is called inverse approach, as it inversely estimates causes of suffer the same fate as the light pen or Microsoft Kinect, namely the movement after having observed the resulting kinematics of motion Most unresolved questions are related to the generation of the neu- projects were closed. capture markers. ral excitation signal. The cooperation with motor control scientists or neuromechanics scientists would be of the largest benefit for us.

PROBLEM STATEMENT AND RESEARCH QUESTION REFERENCES

While inverse biomechanical simulation based on motion capture the design stage, namely they can draw the movement or specify [1] Hoffman, Hunter G., et al. "Virtual reality as an adjunctive pain control during burn [3] Bachynskyi, Myroslav, et al. "Performance and ergonomics of touch surfaces: wound care in adolescent patients." Pain 85.1-2 (2000): 305-309. A comparative study using biomechanical simulation." Proceedings of the 33rd provides valuable outputs covering the whole body to evaluate or alternative targets, observe the outputs and compare them within [2] Banakou, Domna, Parasuram D. Hanumanthu, and Mel Slater. "Virtual embodiment Annual ACM Conference on Human Factors in Computing Systems. ACM, 2015. compare post-desktop interfaces, it still necessitates an experi- minutes rather than weeks. Such research would be of interest for of white people in a black virtual body leads to a sustained reduction in their implicit [4] Delp, Scott L., et al. "OpenSim: open-source software to create and analyze dynamic ment with recording of particular movements involved in the in- all entities involved with developing of VR applications, post-desk- racial bias." Frontiers in human neuroscience 10 (2016): 601. simulations of movement." IEEE transactions on biomedical engineering 54.11 (2007): 1940-1950 teraction, it is complex to apply and the computations consume a top interfaces, gestural interfaces, etc. The method could have ap- large amount of computational resources and time. Our objective plications also in other fields, for example sport trainers or doctors is to develop a fast and reliable method, which can generate the hu- can apply the method to identify optimal movement schemes and man-like movement and provide all biomechanical variables using control for various sports, or can compare the movement of the only the specification of the movement task as an input. sportsmen (analyzed using the inverse approach) to the optimal one and adjust the training accordingly. Such method would allow the designers of post-desktop interfaces The key research questions are related to the motor control, name- to perform an efficient evaluation of the movement ergonomics on ly which neural excitation signals are normally generated and sent

132 | SYMPOSIUM SYMPOSIUM | 133 ABSTRACTS analysis which are employed for sensitive user groups due to load-aware system is proposed (RQ4). This system will able to SYMPOSIUM ethical reasons. evaluate (a) eye behavior, (b) neuronal data, and (c) behavioral 3. Augmenting Cognitive Perception: Based on the outcomes analytics to augment cognitive resources. from the previous studies, an implementation of a work-

RELATED WORK

Lotte et al. [8] and Duchowski [1] have conducted important re- son-dependent features. Such mental alleviation can also be pro- Amplifying Cognitive Performance through search in the domain of feature detection for eye behavior and vided for people with cognitive deficiencies to amplify their overall EEG data. By applying machine learning methods, features can be performance. 33 Workload-Aware User Interfaces tailored for an individual person to increase the accuracy for per- PRELIMINARY RESULTS

Thomas Kosch To evaluate the impact of in-situ assistance, we found that visual ent task complexities [6]. We investigated how eye movements are feedback is most favored compared to auditory or tactile feedback affected under different levels of cognitive workload [4]. Additionally, LMU Munich, Human-Centered Ubiquitous Media [3]. In a follow-up study, we measure EEG data and compare our we have conducted qualitative inquiries at sheltered living organi- [email protected] visual approach with paper instructions within an assembly task. We zations to investigate possible workload alleviations for people with Prof. Dr. Albrecht Schmidt found that cognitive capacity is alleviated when using in-situ projec- mental deficiencies [5]. From observations and inquiries, we propose tions on a neuronal basis [7]. We show how EEG data can be visual- design implications on how workload-aware user interfaces have to ized for users which are not proficient with EEG data [2]. We have be employed. conducted a study which measures the pupil dilation during differ- INTRODUCTION

Cognitive capacities are a scarce resource in everyday life. Current In contrast to these self-rated metrics, physiological measures have variations of mental demand depend often on the type of the task employed as an alternative to quantify cognitive workload on a and its complexity. A key aspect in the research domain of Hu- measurable basis. Physiological measures comprise the measure- PLANNED NEXT STEPS man-Computer Interaction (HCI) is the design of user interfaces ment of biosignals, such as eye tracking or electroencephalography which minimize the cognitive demand and support users through (EEG), which facilitate the understanding of the individual during The next step includes the deployment of the approaches in a re- ies from sheltered work and living facilities will serve as a basis for their task. Cognitive aspects of user interfaces are traditionally eval- times of cognitive strains during an interaction. Integrating work- al-world system. The methodology used in the previous EEG and eye an adaptive cooking system for people with cognitive impairments. uated through questionnaires or the think-aloud protocol. However, load-awareness as additional context-related factor enables future tracking studies will be used to build an adaptive speedreading appli- Furthermore, an interdisciplinary literature research is in progress, these approaches are (a) prone to subjective perception, (b) requires user interfaces to adapt on the current level of cognitive workload cation which adjusts the text visualization and reading speed based where we critically reflect upon ethical aspects about systems which participants to remember their past task performance of a study, from the user. Furthermore, designers gain a better understanding on the cognitive demand. The outcomes from the qualitative inquir- know about cognitive states. and (c) do not provide insights for cognitive workload in real-time. regarding the cognitive demand within their user interface design. OPEN ISSUES RESEARCH QUESTION Integrating workload measures in the real world is still a challeng- ternally or made available in a ubiquitous way (e.g. integrated into This dissertation discusses the key aspects of cognitive workload imagery). RQ2 discusses how a system should react when cognitive ing task. Physiological measures are often prone to external factors, smart garments) to increase the user acceptance of wearing such which arise during interaction with user interfaces. This includes the workload is detected. Past research has used intervention systems which can be compensated by combining sensors or restricting the sensors. following research questions: or provided additional user interface elements for users. However, it user interaction space. Furthermore, sensors must be deployed ex- • RQ1: How can be cognitive workload defined in the domain of HCI is unknown if these elements provide cognitive alleviation. To com- and what are past methodologies to measure cognitive workload? plement this, RQ3 copes with the detection of cognitive workload • RQ2: How can workload-aware interfaces provide suitable adaptiv- through subjective perception, physiological data, or user behavior. REFERENCES ity for different individuals? To detect cognitive workload on a physiological basis, monitoring eye • RQ3: Which physiological sensing modalities provide a reliable behavior and neuronal activity through electroencephalography are [1] Duchowski, Andrew T. "Eye tracking methodology." Theory and practice 328 (2007). [5] Kosch, Thomas, et al. "Smart Kitchens for People with Cognitive Impairments: measure of cognitive workload? employed. RQ4 shows how the findings from the previous research [2] Kosch, Thomas, Mariam Hassib, and Albrecht Schmidt. "The brain matters: a 3d A Qualitative Study of Design Requirements." Proceedings of the 2018 CHI Conference on real-time visualization to examine brain source activation leveraging neurofeedback." Human Factors in Computing Systems. ACM, 2018. • RQ4: How can real-time adaptivity be employed in real-world sce- questions can be incorporated into real-world scenarios. The out- Proceedings of the 2016 CHI Conference Extended Abstracts on Human Factors in Comput- [6] Kosch, Thomas, et al. "Look into My Eyes: Using Pupil Dilation to Estimate Mental narios? comes of these research questions will provide how workload-aware ing Systems. ACM, 2016. Workload for Task Complexity Adaptation." Extended Abstracts of the 2018 CHI Confer- user interfaces can be designed for different user groups and work- [3] Kosch, Thomas, et al. "Comparing tactile, auditory, and visual assembly error-feedback ence on Human Factors in Computing Systems. ACM, 2018. for workers with cognitive impairments." Proceedings of the 18th International ACM [7] Kosch, Thomas, et al. "Identifying Cognitive Assistance with Mobile RQ1 intends to find a set of important cognitive workload models load models. Design guidelines on how to adapt tasks to ensure a SIGACCESS Conference on Computers and Accessibility. ACM, 2016. Electroencephalography: A Case Study with In-Situ Projections for Manual Assembly." for HCI. Current controversial discussions in the field of psychology high user engagement and a machine learning toolbox for eye track- [4] Kosch, Thomas, et al. "Your Eyes Tell: Leveraging Smooth Pursuit for Assessing Cogni- Proceedings of the ACM on Human-Computer Interaction 2.EICS (2018): 11. show that a diverse set of mental workload models are responsible ing and EEG data complement this. tive Workload." Proceedings of the 2018 CHI Conference on Human Factors in Computing [8] Lotte, Fabien, et al. "A review of classification algorithms for EEG-based brain–comput- Systems. ACM, 2018. er interfaces: a 10 year update." Journal of neural engineering 15.3 (2018): 031005. for experiencing cognitive workload (e.g. working memory or visual

RESEARCH APPROACH AND METHODOLOGY

The research approach employs qualitative and quantitative tion and to avoid false positives during user interaction. People methods to answer the stated research questions. This defines with cognitive impairments are also included as a user group three sections: since they benefit from cognitive support during daily life.

1. Understanding People: This part refers to the research ques- 2. Quantifying Cognitive Workload: Possible measures and met- tions RQ1 and RQ2. It is important to understand how people rics of cognitive workload are outlined (RQ3). The used method- behave under cognitive workload to provide suitable interven- ology consists of a combination of quantitative and qualitative

134 | SYMPOSIUM SYMPOSIUM | 135 SYMPOSIUM ABSTRACTS RESEARCH APPROACH AND METHODOLOGY

Incentivized experiments to study decisions with real externalities personality traits) of our participants and vary their decision situa- We study the acceptance of decision randomization in incentivized tion in a controlled manner. This provides us with the opportunity laboratory experiments. Moral dilemmas are created by inducing to compare our results with those of framed questions concerning decisions with negative consequences concerning others as well as preferences for algorithm programming in hypothetical settings. Behavioral Implications of Human-Machine- the decision-makers themselves. We observe individual factors (e.g. Interactions – Experimental Evidence on PRELIMINARY RESULTS Decision Randomization in Ethical Dilemmas We implemented first pilot sessions with a student sample at the We find that decision randomization for moral dilemmas through 34 TUM School of Management’s laboratory for experimental research the use of an algorithm is chosen by about half of the participants in economics. in our pilot settings. Anja Bodenschatz Technical University of Munich, School of Governance, Junior Research Group “Ethics of Digitization” [email protected] PLANNED NEXT STEPS Dr. Matthias Uhl We will now start to vary the characteristics of the moral dilemma analyze individual as well as aggregated decision patterns in connec- situations that our participants face in the laboratory. We will then tion to personal and situational factors.

INTRODUCTION OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Moral decision making in the age of algorithms behavior. When moral reasoning is triggered, people tend to apply Autonomous driving fuels a debate (also present in other contexts utilitarian arguments, but when moral emotions are triggered, peo- We aim to discuss possible external implications of our study design ested to cooperate with other researchers to create and implement of algorithm design, e.g. HR recruiting) on the morality of weighing ple rather react in line with a deontological approach (Lanteri et al., with the ZD.B experts and audience. We would be interested in ex- these field experiments. negative externalities in the context of machine learning - which 2008). Concerning the example of autonomous driving, although tending our research methodology to the field and would be inter- may even include harm of human life. Many decisions that were individual factors of potential accident victims (e.g. young age) trig- previously left to humans are now taken by algorithms. These al- ger strong heuristics (Bleske-Rechek et al., 2010), people might re- ready include moral decisions (Bertelsmann Stiftung, 2018). frain from entering individual factors into algorithmic calculations. REFERENCES Algorithm programming is rule based. This means, it has to be de- The Ethics Commission of the German Federal Ministry of Trans- cided a priori (i.e. before the dilemma materializes) which outcome port and Digital Infrastructure already ruled out any discrimination [1] Bartling, B., & Fischbacher, U. (2011). Shifting the blame: On delegation and responsi- [5] Ethics Commission. (2017). Automated and connected driving. Technical report, Feder- variable an algorithm should maximize. By demanding for an ex- by personal factors in its technical report for the programming of bility. The Review of Economic Studies, 79(1), 67-87. al Ministry of Transport and Digital Infrastructure. https://www.bmvi.de/SharedDocs/ [2] Bertelsmann Stiftung. Was Deutschland über Algorithmen weiß und denkt. EN/PressRelease/2017/084-ethic-commission-report-automated-driving.html plicit commitment to an ethically preferred position, this offers the autonomous cars (2017, p.11): “In the event of unavoidable acci- https://www.bertelsmann-stiftung.de/de/publikationen/publikation/did/was- [6] Lanteri, A., Chelini, C., & Rizzello, S. (2008). An experimental investigation of emotions opportunity to clarify moral orientations in a society. dent situations, any distinction between individuals based on per- deutschland-ueber-algorithmen-weiss-und-denkt/. 2018 and reasoning in the trolley problem. Journal of Business Ethics, 83(4), 789-804. sonal features (age, gender, physical or mental constitution) is im- [3] Bleske-Rechek, A., Nelson, L. A., Baker, J. P., Remiker, M. W., & Brandt, S. J. (2010). [7] Reynolds, S. J. (2006). Moral awareness and ethical predispositions: investigating the Evolution and the trolley problem: people save five over one unless the one is young, role of individual differences in the recognition of moral issues. Journal of Applied No consensus on the answer to ethical dilemmas permissible.” genetically related, or a romantic partner. Journal of Social, Evolutionary, and Cultural Psychology, 91(1), 233-243. For many ethical dilemmas no consensus will be reached in soci- Psychology, 4(3), 115-127. [8] Weber, J. M., Kopelman, S., & Messick, D. M. (2004). A conceptual review of decision [4] Burri, T. (2017). Machine Learning and the Law: 5 Theses. making in social dilemmas: Applying a logic of appropriateness. Personality and Social etal discussions, since by definition they do not have a clear mor- Decision Randomization http://www.mlandthelaw.org/papers/burri.pdf Psychology Review, 8(3), 281-30 ally dominant solution. This becomes obvious when philosophical Burri (2017) makes an interesting case by suggesting that decision ideas of outcome-based (utilitarian) or rule-based (deontological) randomization for exactly these ethical dilemmas can be the way ethics are compared (Reynolds, 2006). Since these ethical theories forward in programming machine learning systems. He argues that define different aspects in situations as morally most important, in cases in which society cannot agree upon a morally preferable they often arrive at different suggestions for an ethically preferable option, implementing a randomization of choice options within al- action. Even within persons there seems to be no consistent choice gorithms seems reasonable.

PROBLEM STATEMENT AND RESEARCH QUESTION

The aim of our study Academic and practical contribution In a first step, our study aims to analyze, if the acceptance of Our research will directly translate into social implications with decision randomization by algorithms is generally accepted as political connotations. As in the case of autonomous driving, a means to address ethical dilemmas. In a second step, we will many ethical dilemmas occurring in algorithm programming, will identify factors which lead to the delegation of moral tasks to yield no societal or political consensus for a preferred outcome. randomization devices, as well as the restrain to do so. Individu- To research the acceptance of randomizing algorithms is thus of al characteristics of decision makers as well as situational factors high societal relevance. As Burri (2017) suggests, the need for the will be analyzed. programming of a priori decisions in ethical dilemmas is prone to hinder the advancement of automatic systems, which could oth- Our main research questions in this regard are: erwise benefit humanity. Our research has strong practical impli- Is randomization generally accepted as an opportunity to address cations, since it will shed further light on the origins of algorithm ethical dilemmas and applied to make ethical decisions? Does a aversion in the German population (Bertelsmann Stiftung, 2018) delegation to a machine offer the same moral relief as the dele- and possible countermeasures to facilitate the progress of digi- gation to another human being? Which individual and situational tization. Furthermore, this research contributes to the literature factors raise the willingness to delegate a moral decision to a ma- on decision making in social dilemmas (Weber at al., 2004) and chine? decision delegation (Bartling & Fischbacher, 2011).

136 | SYMPOSIUM SYMPOSIUM | 137 SYMPOSIUM ABSTRACTS RELATED WORK

Overall, the topic is located within the research area of business pro- proaches for trace clustering to detect frequent behaviour in busi- cess management (BPM). Extensive work was done regarding the ness processes. The approach is presented on a rather technical and interplay of defined reference process models and the executing abstract level. Its application to the case of identifying process activi- information systems as well the design and execution of workflow ties suitable for automation has to be assessed. systems. Approaches, best practices and practical applications are Bridging the gap between process analysis summarized and discussed by Scheer et. al [2]. Given the fact that Looking outside the area of BPM, an interest approach for the overall the book is almost 15 years old new technologies and techniques as vision of a “self-healing” system is presented by Kephart and Chess and process automation – An evidence-based well as new challenges for organizations are not considered. [4]. They introduce the concept of “autonomic computing” outlining how computers manage and optimize themselves. If and how ap- process automation approach In Process Mining various literature exists for identifying patterns proaches from autonomic computing could be applied to business 35 from process data. Bose and Van der Aalst [3] present various ap- process automation is subject to further research. Matthias Stierle PRELIMINARY RESULTS FAU Erlangen-Nürnberg, Chair of Digital Industrial Service Systems As I am still at the beginning of my dissertation project, my main find- Another finding that is expected in the near future is an analysis of [email protected] ing so far is the relevance of the topic as confirmed by many compa- existing concepts for process automation from the literature in order nies that I have spoken to. to identify shortcomings and gaps. Prof. Dr. Martin Matzner

INTRODUCTION PLANNED NEXT STEPS Information systems have enabled organizations to run their pro- these insights the question that naturally arises is “How do we use cesses digitally. The focus of digital process transformation starting this information to improve our process?”. One popular approach for The next step will be an extensive literature review on the related What is still missing up to this point is pinpointing the exact problem in the 1980s was on supporting operations by enabling decentralized making processes faster, decrease costs and increase throughput is work that builds the base for designing the framework. Further- and therefore determining the requirements for the artefacts of the communication and access to information [5]. However, so to speak process automation. Currently, RPA (Robotic Process Automation) is more, with the kick-off of my “Software Campus” project at DATEV I dissertation. On the one hand, this will be achieved by reviewing ex- as a by-product, massive amounts of evidence about process execu- a technology highly discussed in industry and has been proposed as will present my idea and roadmap for developing the process mining isting literature in order to identify existing approaches and on the tion were gathered. For long, this evidence, which is often referred to a trending research topic [1]. RPA vendors promise the automation technique on September 26, 2018. Finally yet importantly, three fur- other hand by input received from industry partners on their needs as event logs, was not used for analysis purposes beyond operation- of recurring human tasks with the help of their software. Companies ther industry partners showed interest in partnering up and possible and expectations to ensure the relevance of the work. al reporting. Process Mining emerged as a technology filling this gap. engaging in such automation endeavors are asking the question of forms of collaboration are currently assessed. With the capabilities of Process Mining business processes can easily “How do we know where or what to automate in our process?” This be assessed from a holistic perspective. Typical performance indica- suggests the existence of potential synergistic effects between pro- tors are throughput times, idle times, which activities are executed cess automation and process analysis. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION and how often, who is involved in process execution. Having gained

I feel that the BPM toolbox is big but only few concepts have found developing the artefacts. As said, various industry partners have al- broad application in organizations. Therefore, I am very eager to un- ready manifested their interest in the research and hence a collabo- PROBLEM STATEMENT AND RESEARCH QUESTION derstand what it takes for e.g. a technique or method to be adopted ration. Of course, I would be happy to engage with further partners by practitioners and how this influences my design decisions when and in general about any form of input and feedback. Companies struggle with transforming process insights into process Therefore, the following contributions are expected to be made: improvements. Process automation is a promising approach for pro- • A framework positioning key concepts and requirements for evi- cess improvement but what to automate is often unclear. dence-based process automation REFERENCES • A method describing the steps for executing an evidence-based Missing out on process automation is hindering organizations from process automat project [1] van der Aalst, W.M.P., Bichler, M. & Heinzl, A. Robotic Process Automation. Bus Inf Syst [4] Kephart, J. O., D. M. Chess. The vision of autonomic computing. Computer 1, 41-50. increasing efficiency and hence from staying competitive. • A process mining technique capable of identifying automation po- Eng, 60(4), 269-272. 2018. 2003. tential from event logs [2] Scheer, A. W., Abolhassan, F., Jost, W., & Kirchmer, M. Business process automation. [5] Mukherji, A. The evolution of information systems: their impact on organizations and Heidelberg: Springer, 2004. structures. Management Decision, 40(5), 497-507. 2002. This leads to the following research questions: [3] Bose, R. J. C., van der Aalst, W. M.P. Context aware trace clustering: Towards improving • Which indicators are suitable for assessing the “automatability” of Practitioners are provided with a toolbox for initiating and executing process mining results. Proceedings of the 2009 SIAM International Conference on process activities? process automation projects based on historic process data. Data Mining. Society for Industrial and Applied Mathematics, 2009. • How can Process Mining be used to determine such indicators? • Which data is required and how can it be collected? Research so far focuses on the interplay between process modelling • How can the insights be used as input for automation? and process automation. This work will extend this view by outlining the capabilities of process analysis for process automation.

RESEARCH APPROACH AND METHODOLOGY

The research will follow the design science research (DSR) paradigm. The research will be conducted in close cooperation with industry After identification of the requirements, various artefacts (frame- partners. Part of the research is done in collaboration with DATEV as work, method, technique) will be built and evaluated. There will be part of the “Software Campus” project. Further collaborations with both qualitative (interviews) as well quantitative (data analysis) eval- industry partners are currently being initiated. uations for the artefacts.

138 | SYMPOSIUM SYMPOSIUM | 139 ABSTRACTS knowledge, this is the first MR-application developed to cue AD-pa- SYMPOSIUM tients in a multi-step daily task. The missing healthy control group and potential carry-over effects in the chosen cross-over design can st be considered as potential limitations of our study. Yet, the applied Fig. 2: 1 person perspective with given virtual step by step methods incorporating both, qualitative and quantitative data, al- guidance lowed for taking the users´ needs into account from the very beginning and enabled us to get a first understanding of the usability “Mixed Reality” as a new therapeutic and feasibility of our novel system. approach to support activities of daily living in patients with chronic neurological disease PRELIMINARY RESULTS 36 Within the pilot study, three patients failed (p=.25) in the MR condi- in patients with stronger needs of support, concluding that the pre- tion and trial durations were sig. longer than in the natural condi- sented “step by step” approach may be most beneficial for more tion (77s±23s vs. 111±24s, P=.02). We assumed that the MR appli- severely affected patients. 90% of patients were able to control the Nina Rohrbach cation serves as a secondary task requesting its own resources and application, and the acceptability appeared to be high, as a large Poster impairing the ADL performance; i.e. the unnatural interaction with part of participants revealed a positive attitude towards the system, Technical University of Munich, Department of Sport and Health Sciences, Human Movement Science Award the application and a drag of attention to the holograms from the although the hardware was considered the main impediment with [email protected] 2. PREIS real objects. However, a model of multiple linear regression sug- 60% of the patients being unsatisfied. gests that the given assistance compensated these additional costs Prof. Dr. Joachim Hermsdörfer 1000€

INTRODUCTION PLANNED NEXT STEPS

With the increasing number of people suffering from stroke and beyond what seems possible with conventional therapy approach- The pilot study indicates that the paradigm of augmented support is with in-depth understanding of the advantages and barriers of our neurodegenerative diseases (e.g. dementia), there is an urgent es. Despite these promising potentials, research is required to un- promising, but the different cues (e.g. number or mode) need to be new approach. need for effective treatment strategies. Innovative digital solutions, derstand how systems can exploit key motor learning principles in evaluated against each other to investigate the most beneficial way e.g. assistive technology devices or Mixed Reality (MR) systems, can order to develop and evaluate evidence-based systems that proof augmenting feedback to the real world environment in people Learning from our pilot results we plan to investigate a larger study benefit patients and the entire healthcare system in various ways, mote functional recovery from neurological impairments. with limited cognitive performance (RQ-2). To investigate the most sample (N=20) and include a matching healthy control group (N=20). beneficial cue in apraxia patients we anticipate, that translating a Suggesting that the presentation of a holographic tool will provide well-recognized clinical apraxia screening tool (pantomiming of ac- more salient information in how to utilize a specific tool successfully, PROBLEM STATEMENT AND RESEARCH QUESTION tions, e.g. pounding in a nail with a hammer) into a holographic set- we expect the strongest differences for the conditions of a dynamic up, will allow us to compare two different characteristics (static vs. 3D cue (e.g., moving holographic hammer) and a static 2D cue (pic- dynamic) in two different dimensions (2D vs. 3D). The combination ture of a hammer) regarding the hand kinematics and the execution A significant proportion of patients suffering from stroke or assistance from caregivers, who guide them through the task and of quantitative data (using motion capture analysis) and qualitative of errors. Alzheimer´s Disease (AD) are symptomatic of apraxia, a neuropsy- prompt them when necessary [1, 3]. data (behavioral observation using video recordings) will provide us chological deficit that impairs the cognitive abilities to carry out activities of daily living (ADL), e.g. preparing a cup of tea (Fig.1) or The inability to live an independent life impacts the patients, their using tools correctly (e.g. hammering) [1, 2]. families, and the healthcare system. Thus, finding solutions to maintain independency is of highest importance [4]. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION MR technology allows for the pursuit of new pathways to study new approaches for the treatment of apraxia. Due to the salient charac- With respect to RQ-1 the implemented application relied on a step or frustrating the user. In general, cooperation in the fields of com- ter of three-dimensional holographic objects it is hypothesized that by step guiding system which was insufficient as patients got con- putational science and machine learning are of high value to assist in Fig. 1: Patient with Alzheimer's the integration into a real ADL scenario provides for entirely new fused due to missing details. A remaining challenge is to develop programing and animating MR-scenarios for both, therapeutic appli- Disease failing in the tea making task opportunities to guide patients through complex tasks by cuing the action recognition models in order to only prompt actions to either cations and assistive systems. desired motor plan. Key research questions (RQ) are: correct the sequence, or prevent further errors, without distracting

1. Is a MR-based support system feasible and usable in patients suffering from apraxia? 2. Which is the most beneficial (holographic) cue to reinforce the REFERENCES As a result of insufficient evidence related to the effectiveness of appropriate motor plan? [1] Bieńkiewicz, M., et al., The tool in the brain: apraxia in ADL. Behavioral and neurological [4] Hermsdörfer, J., et al. Cogwatch–automated assistance and rehabilitation of stroke-in- traditional therapeutic interventions for apraxia, patients rely on correlates of apraxia in daily living. Frontiers in psychology, 2014. 5: p. 353. duced action disorders in the home environment. in International Conference on Engineer- [2] Sperber, C., et al., Apraxia of object-related action does not depend on visual feedback. ing Psychology and Cognitive Ergonomics. 2013. Springer. Cortex, 2018. 99: p. 103-117. [3] West, C., et al., Interventions for motor apraxia following stroke. Cochrane Database Syst RESEARCH APPROACH AND METHODOLOGY Rev, 2008. 1: p. C0004132.

Step 1 – App Development Step 2 – Pilot Testing

An iterative co-design process involving 29 relevant stakeholder, i.e. For feasibility and usability testing (RQ-1), we investigated the dai- AD patients (n=19), family members (n=3), clinicians (n=7), served to ly task of tea making within a randomized crossover study on ten develop a first prototype ADL support system. The developed MR- AD-patients, with each one trial of a natural execution and the MR based application “Therapy Lens” is now available for free down- condition, using a head mounted Microsoft HoloLens, followed by load on the Microsoft store. semi-structured interviews. In the MR condition (Fig. 2), patients were guided through each necessary step giving multidimension- al cues (audio, text, & holographic animations). To the best of our

140 | SYMPOSIUM SYMPOSIUM | 141 SYMPOSIUM ABSTRACTS PRELIMINARY RESULTS

The research has not been started yet.

PLANNED NEXT STEPS

Digitalization in clinical environments After a systematic literature study, a long-term stay at a hospital is in preparation to get to know the different staff groups and their points 37 using graph databases of contact with different patient-centered data sources. OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Oliver Haas There are some vendors of clinical information systems. How are Clinics play a major part in the research questions, so a cooperation FAU Erlangen-Nürnberg, Fakultät Informatik · OTH Amberg-Weiden, Fakultät Wirtschaftsingenieurwesen these systems used? Do they improve the clinical workflow? This of- with a clinical environment is a must and will bring helpful insights [email protected] fers much room for cooperation with companies. to their workflow. Prof. Dr.-Ing. habil. Andreas Maier · Prof. Dr.-Ing. Eva Rothgang

INTRODUCTION

In clinical environments, a lot of heterogenous patientcentered data most cases patient-centered. Can this data be efficiently stored in a is collected. The underlying information is highly connected and in database and used by the staff?

PROBLEM STATEMENT AND RESEARCH QUESTION

Various data sources exist in a clinical environment, e.g. the HIS Interesting questions that arise are e.g. if the clinical workflow can (Hospital Information System) or the RIS (Radiology Information be improved by giving the staff access to a unified source of infor- System). These contain different data points coming from different mation about their patients. What kind of questions do different points in time but describing the same patients. Other interesting groups have? How can the usability for the staff be improved? sources of information are symptoms and treatments of patients. On a more general side: can such a system help answer questions These heterogenous data sources can be unified using graph data- like “Given a patient’s symptoms, are there already treatment paths bases, which are databases that describe their contents as different from similar patients?” or “What treatment paths are most likely nodes (possibly of different types) and relationships between these chosen based on different diagnostic image techniques?”. This can nodes. be achieved by opening the unified graph database to a data mining environment

RESEARCH APPROACH AND METHODOLOGY

Usability studies of clinical databases and their access will be helpful data sources and to different parameters (e.g. treatment path) will indicators. Performance indicators of clinical workflows will have to bring helpful insight to the clinic. be discussed with clinical staff. Their change in regard to different

142 | SYMPOSIUM SYMPOSIUM | 143 ABSTRACTS package, e.g., by typing “npm install leftpad” (instead of “left-pad”), form about the study, but abort the installation of the package. An SYMPOSIUM he installs the adversary’s packages and runs the malware. error message is shown, indicating that the “adversary” mis-typed when programming the “malicous” package, e.g., “install_rotkit” in- To measure the impact of this attack, a two-part empirical study is stead of “install_rootkit.” conducted. In the first part, we uploaded many fake packages with “typosquatted” names to the NPM registry and to package registries A different problem exists with the popular Google Analytics JavaS- of other programming languages. These packages do not contain cript library, which is used to track one’s website visitors. In the Eu- malware, but only collect statistics and inform the programmer or ropean Union, users of Google Analytics, i.e., website owners, must On the Influence of the JavaScript system administrator about the study and his mistake when install- enable the “Anonymize IP” (AIP) extension to conform to the data ing a package. protection laws. Enabling this option truncates the website visitor’s IP Scripting Language and its Ecosystem address before it is submitted to Google. A common mistake is to ei- In the second part, after the first part of the study gained the atten- ther not enable AIP or to send the tracking data to Google before the on Security and Privacy tion of the repository maintainers, we again upload packages with AIP option is set. The library requires the command to enable AIP to “typosquatted” names. In contrast to the first part, we upload them be executed before the tracking command. It does not warn the user 38 under cover to measure the behavior of the maintainers when they if the AIP option is set afterwards. To study how widespread these detect such packages. The “typosquatted” packages contain special- two problems are (not enabled and wrong order of commands), we ly crafted links to other “typosquatted” packages, e.g., by uploading scanned the Internet with an instrumented browser. The dataset Henning Pridöhl them with the same user or IP address. We expect the maintainers consists of about 32,000 Domains ending with “.de” to make sure to look for other “typosquatted” packages when they find one. Again, they address European (or German) visitors. Those domains were University of Bamberg, Privacy and Security in Information Systems Group the packages do not contain malware, but this time they do not in- randomly extracted from various pages of the German Wikipedia. [email protected] · https://www.uni-bamberg.de/psi/ Prof. Dr. Dominik Herrmann PRELIMINARY RESULTS

The first study shows that “Package Typosquatting” is a problem that second part of the study is ongoing, but as of now, about 50% of the INTRODUCTION affects many users. In the course of the study, about 17,000 com- “typosquatted” packages were already deleted. For Google Analytics, puters installed “typosquatted” packages and could therefore be 2,791 of the 32,782 websites do not anonymize the IP address of Popular sites and web apps, e.g., Google Docs and Slack, make tation and tutorials for programmers is discussed [4]. However, the turned into a botnet. About half of the installations were executed their visitors, whereas 650 fail to enable “Anonymize IP” before send- heavy use of the JavaScript scripting language. As a fact, JavaScript is discussion is limited to problems that target security questions di- with full administrative rights. Among the affected hosts, there were ing the tracking data to Google. used on most websites today. Given its prevalence, only very little is rectly; it does not cover general code examples where insecure prac- computers of the government and military of the United States. The known about the extent of security and privacy problems that result tices are promoted and do not address JavaScript specifically. from unintentional programming mistakes in JavaScript or acciden- tal misuse of tools in the JavaScript ecosystem. The issue of program- Moreover, a programming language’s ecosystem and its tools can ming mistakes has mostly been addressed from the perspective of lead to security issues if they are prone to misuse or promote bad XSS vulnerabilities on the client-side, which allow attackers to steal practices. To the best of the author’s knowledge, the only security PLANNED NEXT STEPS users’ data or perform actions on behalf of the users. Furthermore, problem in the ecosystem that was described deals with outdated existing research focuses on detecting those vulnerabilities reliably dependencies. A significant number of websites depend on oldli- The next steps include evaluating the second part of the Ty- addition, website owners which implement Google Analytics incor- and at a large scale [1], rather than asking what caused the program- brary versions with security issues [5]. Understanding the deficits in posquatting and finding more common problems on client-side rectly could be asked if they have reason or ideas why they made mer’s mistake in the first place. It has been argued, that the usability the JavaScript language, its APIs, documentation, and ecosystem that and server-side JavaScript that can be evaluated systematically. In the mistake. of a programming language and its APIs influences the security of facilitate security-related mistakes, helps to propose improvements. computer programs [2,3]. However, no such analysis of JavaScript’s Thus, it may enhance the security of software written in JavaScript language constructs and APIs, including those of commonly used and consequently the security of many websites and web apps. libraries, has been conducted yet. Also, the influence of documen- OPEN ISSUES AND POSSIBILITIES FOR COOPERATION

Currently, it is an open question whether testing JavaScript pro- up is demanding, getting ideas from researchers that perform such grammers, e.g., by providing tasks that are difficult to implement, experiments more often would be useful. PROBLEM STATEMENT AND RESEARCH QUESTION yields results in regard to new problems. Designing a sensible set-

JavaScript heavily depends on external packages, because the lan- the server-side to provide a backend for the frontend. This may intro- guage itself is minimal and does not offer many of the predefined duce new problems. Hence, we ask: What are common problems on functions found in other languages. Therefore, typical JavaScript soft- the client-side and on the server-side and how do they differ? How wide- REFERENCES ware has many dependencies. As described before, those depend- spread are those problems? Why do programmers make those mistakes? [1] Sebastian Lekies et al. 25 Million Flows Later – Large-scale Detection of DOM-based Code Security. In IEEE Security and Privacy (S&P ’16). 2016. encies are often outdated. Therefore, the following question arises: XSS. In Proc .of the 20th ACM Conference on Computer and Communications Security [5] Lauinger et al. Thou Shalt Not Depend on Me: Analysing the Use of Outdated What other common problems exist in the JavaScript‘s ecosystem regard- Being aware of the identified problems, the final goal is always to (CCS ’13). 2013. JavaScript Libraries on the Web. In Proc. of the 24th Annual Network and Distributed ing package management and how do they impact security and privacy? enhance the security. Consequently, we ask: Which improvements can [2] Acar et al. You Are Not Your developer, Either: A Research Agenda for Usable Security System Security Symposium (NDSS ’17). 2017. and Privacy Research Beyond End Users. In IEEE Cybersecurity Development (SecDev [6] Wang et al. Strider Typo-Patrol: Discovery and Analysis of Systematic Typo-Squatting. be proposed to reduce the number of mistakes made by programmers ’16). 2016. In Proc. of the 2nd Workshop on Steps to Reducing Unwanted Traffic on the Internet JavaScript is increasingly not only used on the client-side, but also on and system administrators [3] Green et al. Developers Are Not the Enemy! The need for Usable Security APIs. In IEEE (SRUTI ‘06). 2006. Security & Privacy (S&P ’16). 2016. [4] Acar et al. You Get Where You’re Looking For: The Impact of Information Sources on RESEARCH APPROACH AND METHODOLOGY

To describe a problem with JavaScript’s package management by the adversary or in the worst case, downloads malware. The (NPM Registry), we apply an attack known as “Domain Typosquat- same attack can be applied to JavaScript package names instead ting” [6] to this new domain. In “Domain Typosquatting” an adver- of domains: In the NPM registry, anyone can register a JavaScript sary wants to distribute malware or earn money by showing ads. package under any name that is not already taken. This allows for For this purpose, she registers a domain that is quite similar to an an attack we call “Package Typosquatting.” Theadversary uploads existing, popular domain, e.g., “gogle.com,” which is a common- a new package that contains malware. For the package name, he ly mis-typed version of “google.com.” If a user now accidentally uses a name that is similar to an already existing, popular pack- mis-types “gogle.com” into her browser, she is displayed with ads age. If a programmer or system administrator now installs the

144 | SYMPOSIUM SYMPOSIUM | 145 ABSTRACTS mance of the developed model highly depends on the amount also shift input parameters to more accessible data in typical bat- SYMPOSIUM and quality of the available training data. For that reason a main tery pack. research goal is to not only predict an accurate battery’s SOH, but

PRELIMINARY RESULTS

The communication between inverter and battery was set up and a cords contains only current and voltage samples at a frequency of 10 few automated charge and discharge cycles were successfully com- kHz. The following test cycles in the near future will be recorded with State-of-health estimation pleted. Even some pulse tests for characterization were recorded a higher frequency up to 144 kHz, together with further information and observed. The hence resulted calculation of the internal resist- e.g. the temperature. 39 of lithium ion batteries ance of the battery provides plausible outcomes. However those re- Alexander Wallis PLANNED NEXT STEPS University of Applied Science Landshut Beside the ongoing data collection, next steps in the project involves ed and compared with each other. To finalize the research, the de- [email protected] developing a faster parameter identification for the battery under veloped prediction model will be examined. Most of the time ma- Prof. Dr. Sascha Hauke test. That includes a well-structured dataset with accurately defined chine learning models are working as black box models – they do explanatory variables. This data preprocessing is one of the primary not contribute to an improved understanding of the electrochemical task in machine learning and requires the most time of the devel- processes within the battery. A following feature selection, e.g. prin- oping process. Afterwards the developing of a first prediction mod- ciple component analysis (PCA), might provide new insights in the INTRODUCTION el and the comparison of the results with SOH estimation methods complex ageing mechanism of Li-ion batteries and might offer possi- from related work is the aspiration. In addition different machine bilities for a more efficient battery usage. learning models and varying learning approaches will be implement- With the rising popularity of electric vehicles (EVs), efficient use of and degeneration mechanism in a battery is the main and most chal- power battery packs is a major topic of interest. Due to the improve- lenging goal [2]. The SOH determines the health of the battery and it ments of lithium ion (Li-ion) technology in terms of reliability and compares the battery's current state variables to their original rated safety over the past forty years, Li-ion based batteries have become values. To assure an efficient and economic use in EVs, it is widely OPEN ISSUES AND POSSIBILITIES FOR COOPERATION the preferred choice for EVs [1]. Li-ion batteries are complex systems accepted that a battery reached its end-of-life (EOL) for vehicle appli- and their performance degrades with time and usage. Even though cation when the SOH decreases to 80% [3–5]. The generalizability of the proposed SOH estimation method needs that reason access to a lager variety of batteries arises the opportu- the state-of-health (SOH) is well defined, the identification of ageing further investigations. Therefor studies with different types of Li-ion nity for a possible cooperation. Furthermore a close dialogue with batteries should be conducted. Especially the examinations of bat- electrochemical experts might give new impetus how to achieve bet- teries at different degeneration stages is a promising approach. For ter estimations of the actual battery’s SOH. PROBLEM STATEMENT AND RESEARCH QUESTION

An estimation of the battery’s current SOH is thus essential not only timation methods in scientific literature suffer from disadvantages REFERENCES to rate the performance during operation but to identify a possible such as extensive laboratory work, need for additional equipment or second-life usage. Since Li-ion batteries are an expensive component permanent damage [8]. For that reason our research focuses on a [1] Schalkwijk, W. and Scrosati, B. Advances in lithium-ion batteries. Kluwer Academic/ [7] Neubauer J, Pesaran A. PHEV/EV Li-ion battery second-use project; 2010. Plenum Publishers, New York, 2005. [8] Klass, V., Behm, M. and Lindbergh, G. A support vector machine-based state-of-health in EVs, second-life applications are therefore a promising market [6], novel method that achieve battery performance estimation without [2] Balagopal, B. and Chow, M. The state of the art approaches to estimate the state of estimation method for lithium-ion batteries under electric vehicle operation. Journal e.g. vehicle to grid applications, grid-based stationary applications, prior knowledge or experimental preparations. A purely data-driven health (SOH) and state of function (SOF) of lithium Ion batteries. 2015 IEEE 13th INDIN, of Power Sources, 270 262-272, 2014. off-grid-based stationary applications and mobile energy storage ap- approach based on charge and discharge behavior of batteries and 2015. [9] Huynh, P. Beitrag zur Bewertung des Gesundheitszustands von Traktionsbatterien in [3] Diao, W., Jiang, J., Zhang, C., Liang, H. and Pecht, M. Energy state of health estimation Elektrofahrzeugen, 2016. plications [7]. To characterize second-life batteries, SOH estimation a possible deeper understanding of ageing factors of Li-ion batteries for battery packs based on the degradation and inconsistency. Energy Procedia, [10] Pyne, M., Yurkovich, B. and Yurkovich, S. Capacity fade estimation using supervised plays an important role to guarantee a safe and reliable system’s is the main goal. 142 3578-3583, 2017. learning. 2017 IEEE Conference on Control Technology and Applications (CCTA), 2017. operation. Without prior knowledge of their usage history, SOH es- [4] Klass, V. Battery Health Estimation in Electric Vehicles. KTH Royal Institute of Technolo- [11] Idaho, T. & National, E. U. S. Department of Energy Vehicle Technologies Program gy, Stockholm, 2015. Battery Test Manual for Plug-In Hybrid Electric Vehicle, 2008. [5] USABC. Electric vehicle battery test procedures manual-revision 2; 1996. [12] Battery test manual for plug-in hybrid electric vehicles, INL/EXT-07-12536, 2008. [6] Tong, S., Same, A., Kootstra, M. and Park, J. Off-grid photovoltaic vehicle charge using second life lithium batteries: An experimental and numerical investigation. Applied RESEARCH APPROACH AND METHODOLOGY Energy, 104 740-750, 2013.

Although statistical learning methods like artificial neural net- SOC for resistance determination [11, 12]. Both related works are works (ANNs) or support vector machines (SVMs) have been suc- focusing on online estimation methods, this basically means that cessfully applied for state-of-charge (SOC) estimation, purely da- the SOH is estimated under vehicle working condition. Those ma- ta-driven methods for SOH estimations are rather uncommon in chine learning models for online estimations are therefore limited literature [8]. Most of them are focusing on the accuracy of cell’s in terms of memory and computational power. In this research SOH, rather than the SOH of battery packs, which are not directly project stationary battery packs are used, so the data collection is equivalent [3]. Two machine learning approaches for SOH estima- not limited to the sensors available in EVs and also more complex tion are introduced in [4, 9]. In [4] a SVM is used for predicting prediction models can be developed. But analogous to the previ- the SOH contrary to the linear regression in [9]. Even though the ously mentioned dissertations, a test cycle for battery characteri- presented learning models were working with different machine zation and parameter identification needs to be designed. The de- learning approaches, they are both using the capacity and inter- veloping process for that can be described by the following steps: nal resistance as primary degeneration indicators for the battery. • Setting up system infrastructure. Currently, the decrease of capacity and the increase of internal • Gathering battery data. resistance are the main indices to represent the SOH [2, 10]. To • Developing machine learning model. identify those parameters a standard performance test is conducted: two successive low constant-current discharge test for Thereby the actual “modelling” of the prediction model is an iter- capacity determination and four pulse tests at 90, 70, 50, and 30% ative process composed of training and testing, and the perfor-

146 | SYMPOSIUM SYMPOSIUM | 147 SYMPOSIUM ABSTRACTS RELATED WORK

Roland Heuermann et al. [2] illuminate digitization in public admin- sciences and parties, and offers recommendations of action for IT istration on all levels of government. While they sometimes focus on and its role in public administration. economic aspects, the book gives an interesting overview of involved User Centered workflow analysis for innovative Systems in public PLANNED NEXT STEPS

administration One imminent part of my research will revolve around observing and ical workflows. I wish to identify some generic workflows, which are categorizing workflows and procedures in public administration. I transferrable to other departments. Additionally I will further reach 40 could form a cooperation with the purchase department of the Uni- out to other departments and public offices of the city of Regensburg versity of Regensburg where I will conduct a first study and be able for possible cooperation. Florian Bockes to gather qualitative data through interviews and observation of typ- University of Regensburg Chair of Media Informatics, Physical Affordances for Digital Media and Workflows [email protected] · https://hci.ur.de/people/florian_bockes OPEN ISSUES AND POSSIBILITIES FOR COOPERATION Dr. Raphael Wimmer Aside from the interdisciplinary approach and above named fields periods – the industrial revolutions – and how problems and fears in additional connecting factors might be found in information ethics, work environments can be overcome [1]. One open issue that can´t work ethics, psychology and anything related to e-government. be grasped yet is the extent of the technological prowess of future INTRODUCTION Ethics and psychology might reveal links to the last society shaping innovations.

Digitization is complex and it will concern every individual of our fordances for Digital Media and Workflows” is participating in the REFERENCES society. If we think about the early stages of the industrialization, realization and implementation of novel hard- and software sys- some similarities regarding the scale of the upcoming challenges tems. With the background of media informatics, I want to develop [1] Heuermann, R., Engel, A., & von Lucke, J. (2018). Digitalisierung: Begriff, Ziele und [3] Lazar, J., Feng, J. H., & Hochheiser, H. (2017). Research methods in human-computer and opportunities emerge and are yet to be fully revealed. Since reliable methods to evaluate the Usability and practical impact on Steuerung. In Digitalisierung in Bund, Ländern und Gemeinden (pp. 9-50). Springer interaction. Morgan Kaufmann. this process has already started and happens faster than some of workflows of those systems. Furthermore, I want to focus my re- Gabler, Berlin, Heidelberg. [2] Heuermann, R., Tomenendal, M., & Bressem, C. (Eds.). (2018). Digitalisierung in us might be able to follow, we need to adapt to this phenomenon, search on the domain of public administration in Germany. I hope Bund, Ländern und Gemeinden: IT-Organisation, Management und Empfehlungen. be flexible for new technologies and overcome gaps between the that the resulting optimizations will benefit not only civil servants Springer-Verlag. so-called digital natives and older generations. but also any citizen who is in need of the services of this governing authority. With my PhD thesis, I want to contribute to solutions, which tackle those challenges. In specific, our Research Group “Physical Af-

PROBLEM STATEMENT AND RESEARCH QUESTION

Since this assignment requires knowledge and thoughtfulness in • What qualities does this tool need in order to be adaptive, easy to many scientific fields beyond media informatics like administration use and scientific valid? informatics, administration science, organization science, socio in- • Is it possible to measure real improvements regarding efficiency, formatics and others, I am still gathering literature in order to tackle effectiveness and satisfaction? this challenge appropriately. Additionally I need to conduct some • How can new Systems be introduced in German authorities, with- first small experiments to get a better grasp of the matter. As a small out any loss in motivation and minimal resistance? outlook, I identified the following questions, which will contribute to • How can we preserve important physical affordances and use their the problem statement of my thesis: advantages in everyday workflows? • Can we offer a generic evaluation tool for workflows and prototypes in German public administration?

RESEARCH APPROACH AND METHODOLOGY

In order to solve the problems discussed above I want to develop expert evaluation, contextual inquiry, questionnaires and others [3]. methods and models, which allow us to evaluate and describe user I can also imagine making some psychophysiological measurements experience of prototypes and workflow innovations in public admin- regarding user engagement and motivation. I believe that it might be istration. With my generic approach, consisting of a combination of of importance to understand how we can increase acceptance and qualitative and quantitative analysis, I hope to make different proto- motivation with novel hard- and software systems in this domain. types comparable and thus make it easier to identify deterioration I expect to contribute to the understanding of workflows in public and improvements. In the first phase of my PhD, I want to try differ- administration and plan to offer guidelines for software developers ent established methods like semi-structured interviews, heuristics, and engineers for further improvement in the digital era.

148 | SYMPOSIUM SYMPOSIUM | 149 IMPRESSUM

1. Auflage, Dezember 2018

Inhaltliche Prof. Dr. Dr. h.c. Manfred Broy – Wissenschaftlicher Geschäftsführer ZD.B Ansprechpartner Dr. Nina Höhne – Koordinatorin ZD.B Graduate Program PD Dr. Daniel Méndez – Koordinator ZD.B Nachwuchsforschungsgruppen/Professuren

Herausgeber Zentrum Digitalisierung.Bayern Lichtenbergstr. 8 85748 Garching +49 (0) 89 / 24 88 071-00 E-Mail: [email protected] www.zentrum-digitalisierung.bayern

Editoren Dr. Nina Höhne PD Dr. Daniel Méndez Jessica Lober Sarah Hölzl Dr. Julia Bartels

ZD.B Symposium 2018 wurde durchgeführt in Kooperation mit:

150 IMPRESSUM | 151