Emergent Computing Joint ERCIM Actions: ‘Beyond the Horizon’ Anticipating Future and Emerging Information Society Technologies

Home , Darenet, License proliferation, Middleware, Programming paradigm, Semantic mapper

European Research Consortium for Informatics and Mathematics Number 64, January 2006 www.ercim.org

Special: Emergent Computing Joint ERCIM Actions: ‘Beyond the Horizon’ Anticipating Future and Emerging Information Society Technologies

European Scene: Routes to Open Access Publishing CONTENTS

JOINT ERCIM ACTIONS THE EUROPEAN SCENE

4 Fourth ERCIM Soft Computing Workshop The Routes to Open Access by Petr Hajek, Institute of Computer Science, Academy of 16 Open Access: An Introduction Sciences / CRCIM, Czech Republic by Keith G Jeffery, Director IT, CCLRC and ERCIM president

4 Cor Baayen Award 2006 18 Publish or Perish — Self-Archive to Flourish: The Green Route to Open Access 5 Second ERCIM Workshop ‘Rapid Integration of Software by Stevan Harnad, University of Southampton, UK Engineering Techniques’ by Nicolas Guelfi, University of Luxembourg 19 The Golden Route to Open Access by Jan Velterop 5 Grid@Asia: European-Asian Cooperation Director of Open Access, Springer in Grid Research and Technology by Bruno Le Dantec , ERCIM Office 20 ERCIM Statement on Open Access

6 GFMICS 2005 — 10th International Workshop on Formal 21 Managing Licenses in an Open Access Community Methods for Industrial Critical Systems by Renato Iannella National ICT Australia by Mieke Massink and Tiziana Margaria 22 W3C at the Forefront of Open Access

Beyond-The-Horizon Project by Rigo Wenning, W3C 7 Bits, Atoms and Genes Beyond the Horizon 23 Cream of Science by Dimitris Plexousakis, ICS-FORTH, Greece by Wouter Mettrop, CWI, The Netherlands

8 Thematic Group 1: Pervasive Computing and Communications SPECIAL THEME: by Alois Ferscha, University of Linz, Austria EMERGENT COMPUTING 24 Introduction to the Special Theme 9 Thematic Group 2: by Heather J. Ruskin and Ray Walshe, Dublin City University/IUA, Nanoelectronics and Nanotechnologies Ireland by Colin Lambert, University of Lancaster, UK Invited article: 9 Thematic Group 3: 26 Swarm-Based Space Exploration Security, Dependability and Trust by Michael G. Hinchey, Roy Sterritt, Christopher A. Rouff, by Michel Riguidel, ENST-Paris, France James L. Rash and Walt Truszkowski

10 Thematic Group 4: Invited article: Bio-ICT Synergies 27 Collective Intelligence and Evolution by Fernando Martin-Sanchez, Institute of Health Carlos III, Madrid by Akira Namatame, Akira Namatame, National Defense Spain Academy, Japan

11 Thematic Group 5: Invited article: Intelligent and Cognitive Systems 28 Evolving Game-Playing Strategies by Rolf Pfeifer and Alois Knoll, University of Zurich, Switzerland with Genetic Programming by Moshe Sipper, Ben-Gurion University, Israel 12 Thematic Group 6: Software Intensive Systems 30 Collaborative Online Development by Martin Wirsing, Universität Munchen, Germany of Modular Intelligent Agents by Ciarán O’Leary, Dublin Institute of Technology, Mark Humphrys 13 Quantum Information Processing and Communication and Ray Walshe, Dublin City University/IUA, Ireland Preparation for FP7 in Future and Emerging Technologies Unit, DG INFSO, European Commission 31 CE-Ants: Ant-Like Agents for Path Management in the Next- Generation Internet NEWS FROM W3C by Otto Wittner and Bjarne E. Helvik, NTNU, Norway

14 The Web on the Move 33 ISEE – A Framework for the Evolution and Analysis of Recurrent Neural Networks 14 W3C to Internationalize and Secure Voice Browsing for Embodied Agents by Martin Hülse, Steffen Wischmann, and Keyan Zahedi, 15 W3C Workshop: ‘Toward a More Secure Web’ Fraunhofer Institute for Autonomous Intelligent Systems – AIS, 15 W3C Seminar on Web Services and Semantics Fraunhofer ICT Group, Germany

15 Latest W3C Recommendations 34 Emergent Intelligence in Competitive Multi-Agent Systems 15 W3C Continues ICANN Participation by Sander M. Bohte, Han La Poutré, CWI, The Netherlands

35 Building a Computational Digital Economy through Interdisciplinary Research by Petros Kavassalis, University of Crete, Greece, and Konstantin Popov, Konstantin Popov SICS, Sweden

37 Models of Multilateral Cooperative Behaviour by Milan Mareš Institute of Information Theory and Automation, Academy of Sciences / CRCIM, Czech Republic

38 Emergent Walking Behaviour in an Aibo Robot by Cecilio Angulo, Ricardo A. Téllez and Diego E. Pardo, Technical University of Catalonia (UPC) / SpaRCIM, Spain

2 ERCIM News No. 64, January 2006 Next issue: April 2006 — Special theme: Space Exploration CONTENTS

R&D AND TECHNOLOGY TRANSFER

39 Concept-Based Text Representations for Categorization Digital Libraries Problems 58 XSMapper: a Service-Oriented Utility for XML Schema by Magnus Sahlgren, SICS, Sweden Transformation by Manuel Llavador and José H. Canós, Technical University of 40 A Framework for Efficient Statistical Modelling Valencia/SpaRCIM, Spain by Daniel Gillblad and Anders Holst, SICS, Sweden Bioinformatics 41 Rules, Inferences and Robust Approximation at Work 59 Analysis and Modelling of Genomic Data by Antonín Dvořák, Vilém Novák, Viktor Pavliska, University of by Anna Tonazzini, Francesco Bonchi, Stefania Gnesi, Ercan Ostrava, Czech Republic Kuruoglu and Sergio Bottini, ISTI-CNR, Italy

42 Particle Swarm Optimization for the Reconstruction of Numerical Mathematics Permittivity Range Profiles from Microwave Measurements 61 Numerical Mathematics Consortium by Simone Genovesi and Emanuele Salerno, ISTI-CNR, Italy Distributed Systems 62 CORRECT – Developing Fault-Tolerant Distributed Systems 44 Building Blocks from Biology for the Design of Algorithms for by Alfredo Capozucca, Barbara Gallina, Nicolas Guelfi, Patrizio the Management of Modern Dynamic Networks Pelliccione and Alexander Romanovsky, University of Luxembourg by Gianni A. Di Caro, Frederick Ducatelle, Luca Maria Gambardella, Andrea Rizzoli Istituto Dalle Molle di Studi Ubiquitous Computing sull'Intelligenza Artificiale (IDSIA), Manno-Lugano, Switzerland 63 The WearIT@work Project: Empowering the Mobile Worker with Wearable Computing 45 Chemical Programming of Self-Organizing Systems by Elisa Basolu, Massimo Busuoli, ENEA - Italian National by Jean-Pierre Banâtre, Pascal Fradet and Yann Radenac, Agency for New Technologies, Energy and the Environment and INRIA/IRISA, France Mike Lawo, University of Bremen, Germany

46 A Development Model Radio Technology for Robust Fault-Tolerant Design 64 PULSERS Delivers on Phase 1 – Europe to Adopt a Ruling for by Andy Tyrrell and Hong Sun, University of York, UK Ultra Wide Band by Walter Hirt, IBM Research GmbH, Switzerland 48 Emergent Properties of the Human Immune Response to HIV Infection: Results from Multi-Agent Computer Simulations Security and Trust by Ashley Callaghan, Dublin City University / IUA, Ireland 66 AVISPA: Automated Validation of Internet Security Protocols and Applications 49 Network Emergence in Immune System Shape by Alessandro Armando, Università di Genova, Italy, David Basin, by Heather Ruskin, and John Burns, Dublin City University, Ireland Luca Viganò, ETH Zurich, Switzerland, Jorge Cuellar, Siemens and Michael Rusinowitch, Laboratoire Lorrain de Recherche en 50 Agent-Based Modelling of Viral Infection by Dimitri Perrin, Dublin City University / IUA, Ireland Informatique et ses Applications, France Pattern Recognition 51 The ‘Decent’ Project: Decentralized Metaheuristics 67 Prague Texture Segmentation Data Generator and Benchmark by Enrique Alba, University of Málaga / SpaRCIM, Spain, by Stanislav Mikeš and Michal Haindl, Institute of Information and Martin Middendorf, University Leipzig, Germany Theory and Automation, Academy of Sciences/CRCIM, Czech

53 Evolutionary Methods make New Effective Laser Shapes Republic by Thomas Bäck and Joost N. Kok, Leiden University, Multimedia The Netherlands 68 Visiting Hybrid Museums: A Colony of Ants in your Pocket by Javier Jaén and José A. Mocholí, Universidad Politécnica de 54 Neural Net Modelling in Financial Engineering of Options Valencia / SpaRCIM, Spain by Jerome Healy, Longchuan Xu, Maurice Dixon, Fang Fang Cai, Brian Read and Brian Eales 70 Edutainment and Game Theory Realization through an Open- Source UNESCO Virtual Perspolis Project 55 Self-Optimization in a Next-Generation Urban Traffic Control by Sepideh Chakaveh, Olaf Geuer, Fraunhofer Institute for Media- Environment Communication, Germany; and Stefan Werning, Massachusettes by Raymond Cunningham, Jim Dowling, Anthony Harrington, Institute of Technology, USA Vinny Reynolds, René Meier and Vinny Cahill, Trinity College Dublin / IUA, Ireland Web Tool 71 CONFIOUS: Conference Management System with 56 Self-Organized Routing in Mobile Intelligence, Power and Style Ad Hoc Networks using SAMPLE by Manos Papagelis and Dimitris Plexousakis, ICS-FORTH, Greece by Jim Dowling and Stefan Weber, Trinity College Dublin / IUA, Ireland 72 Announcement: Call for Take up Actions, Joined sub-Projects to the AXMEDIS project of the European Commission

EVENTS

73 Reports: EU-INDIA’05 - Second Annual Conference of the ICT for EU-India Cross Cultural Dissemination Project by María Alpuente, Santiago Escobar, and Moreno Falaschi, Technical University of Valencia / SpaRCIM, Spain

73 7th Workshop of the ERCIM Working Group ‘Matrix Computations and Statistics’

74 ANNOUNCEMENTS

74 LEGAL NOTICE

75 IN BRIEF

ERCIM News No. 64, January 2006 3 JOINT ERCIM ACTIONS

Call for Candidates Fourth ERCIM Soft Cor Baayen Award 2006 Computing Workshop The Cor Baayen Award, awarded to a most promising young researcher in computer science and applied mathematics, was created in 1995 to honour the first ERCIM President, by Petr Hajek and is open to any young researcher having completed their PhD thesis in one of the ‘ERCIM countries’: Austria, Belgium, Czech Republic, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Luxembourg, Norway, Spain, Sweden, Switzerland, The The fourth workshop of the ERCIM Netherlands and the United Kingdom. Working Group on Soft Computing was held jointly with the conference Logic of The award consists of a cheque for 5000 Euro together with an award certificate. The selected fellow will be invited to the ERCIM meetings in autumn 2006. A short article on the Soft Computing IV, held in Ostrava, winner, together with the list of all candidates nominated, will be published in ERCIM News. Czech Republic on 5-7 October, 2005. The workshop was attended by over 40 Rules for Nomination scientists from nine European countries • Nominees must have carried out their work in one of the 'ERCIM countries': Austria, and provided an excellent platform for Belgium, Czech Republic, Finland, France, Germany, Greece, Hungary, Ireland, Italy, the strengthening of working contacts Luxembourg, Norway, Spain, Sweden, Switzerland, The Netherlands and the United Kingdom and exchange of information. Alongside • Nominees must have been awarded their PhD (or equivalent) not more than two years the 29 papers presented, four speakers prior to the date of nomination gave keynote presentations. They were: • A person can only be nominated once for the Cor Baayen Award. • Francesc Esteva, IIIA, CISC, Spain: ‘Logics of a Continuous T-norm and Submitting a Nomination • Nominations should be made by a staff member of the university or research institute Its Residuum with Truth Constants’ where the nominee is undertaking research. Self nominations are not accepted. • Daniele Mundici, University of • Nominations must be submitted with an online form Florence, Italy: ‘Hajek’s Basic Logic • Alternatively contact the ERCIM Executive Committee member (the national contact and Multi-Channel Games with Lies’ point) for the country in which the nominee is undertaking research • Umberto Straccia, ISTI-CNR, Italy: Deadline ‘Fuzzy Description Logics and the • Nominations must be submitted by 15 April 2006. Semantic Web’ • Dag Westerstahl, Gothenburg Further information can be obtained from your national contact or from the Cor Baayen Award University, Sweden: ‘Generalized coordinator László Monostori, SZTAKI, ([email protected]). Quantifier Theory: an (other) Area http://www.ercim.org/activity/cor-baayen.html where Logic Meets Linguistics and Computer Science’. Result of the 2005 Cor Baayen Award The local organizer was the Institute for Research and Applications of Fuzzy Modelling, at the University of Ostrava. Winner: Link: Milan Vojnovic, Microsoft Research, UK http://www.cs.cas.cz/ercim Finalists: Urtzi Ayesta, France Please contact: Francesco Bonchi, Italy Petr Hajek, Soft Computing WG coordinator Santiago Escobar, Spain Institute of Computer Science, Marc Esteva, Spain Academy of Sciences / CRCIM, Keir Fraser, United Kingdom Czech Republic Milan Vojnovic (right) received the 2005 Emmanuel Frécon, Sweden Tel: +4202 6605 3760 Cor Baayen Award from the ERCIM Petr Gebousky, Czech Republic E-mail: [email protected] president Keith Jeffery at a ceremony Claudio Mattiussi, Switzerland http://www.cs.cas.cz/~hajek during the ERCIM meetings in Louvain- Hervé Rivano, France La-Neuve, Belgium on 26 October 2005. Fabrizio Silvestri, Italy Milan Vojnovic, originally from Croatia Rene Sitters , The Netherlands completed his PhD at Ecole Filip Sroubek, Czech Republic Polytechnique Fédérale de Lausanne François-Xavier Standaert, Belgium (EPFL) in Switzerland and has Martin Svensson, Sweden subsequently worked as an Associate Markku Turunen, Finland Researcher in the Systems and Kuldar Taveter, Finland Networking Group at Microsoft Research Kilian Weniger, Germany in Cambridge. Philipp Woelfel, Germany

4 ERCIM News No. 64, January 2006 JOINT ERCIM ACTIONS

Second ERCIM Workshop Grid@Asia: ‘Rapid Integration of Software European-Asian Cooperation Engineering Techniques’ in Grid Research and Technology by Nicolas Guelfi by Bruno Le Dantec

The Rapid Integration of Software Engineering Techniques ERCIM organises the second Grid@Asia Workshop 20-22 (RISE) Working Group held the second edition of its February 2006 in Shanghai. The workshop is hosted by the international workshop in Heraklion, Crete, Greece on 8-9 Jiao Tong University. Some 100 participants have been September, 2005. Forty papers were submitted to the invited from European and Asian research institutes, industry workshop, of which nineteen were selected. The proceedings and ministries of research with the aim to foster co-operation of RISE are expected to be published in the Springer LNCS with significant Grid actors within the European 6th and 7th series (last year’s volume was No. 3474). Framework Programme (FP).

Six sessions were organized covering many aspects of the inte- The workshop will comprise four sessions: the first on applica- gration of complementary mature software engineering tech- tions where Grid technologies play an essential role such as niques. This year we covered areas such as modelling safety emergency response systems, advanced manufacturing, traffic case evolution; practical approaches to mapping UML to simulation. The second session will be dedicated to joint VHDL; context-aware service composition in pervasive com- research centres involving European and Chinese teams, ie, the puting environments; techniques for representing product line Sino-French Laboratory in Computer Science, Automation and core assets in MDA/PIM for automation; formal development Applied Mathematics (LIAMA), the Sino-German Joint of reactive fault-tolerant systems; stepwise feature introduction Software Institute (JSI) and the Southampton Regional in practice; prototyping domain-specific formal languages; and e-Science Centre. The third session will take stock of the exist- aspects and contracts. ing research projects partially funded by the European Commission and involving Chinese partners. The fourth session will be dedicated to future co-operation emerging from call 5 and 6 of the Information Society Technologies (IST) Programme of FP6.

At the end of the workshop, an ‘Information Day’ will be organised by the European Commission to promote the IST Call 6 to be closed on 25 April 2006. This call covers the topics ‘Ambient Assisted Living (AAL)’, ‘Advanced Robotics’, ‘Search Engines for Audio-Visual Content’ and ‘International Cooperation’. Second ERCIM RISE Workshop. This workshop is the second of a series of strategic workshops in Asia in the frame of the Grid@Asia initiative supported by The keynote talk was given by Bertrand Meyer, Professor of the European Commission. The first event took place from 21 Software Engineering at ETH Zürich, and Founder and Chief to 23 June 2005 in Beijing, hosted by the Beihang University. Architect of Eiffel Software in California. He is the inventor of The aim of the iniative is to expand European Grid expertise the ‘Design by Contract’ method, and author of many books, through co-operation with leading Asian and international including ‘Object-Oriented Software Construction’, ‘Reusable teams in the field. Software’, ‘Introduction to the Theory of Programming Languages’ and ‘Eiffel: The Language’. Both his talk and the More information: http://www.gridatasia.net discussion it generated were very productive for the workshop.

Last but not least, we thank our Greek partner Anthony Savidis from FORTH, who organized our wonderful stay in Crete.

Link: RISE homepage: http://rise.uni.lu

Please contact: Nicolas Guelfi, RISE Chairman, University of Luxembourg Tel: +352 420 101 251, E-mail: [email protected]

ERCIM News No. 64, January 2006 5 JOINT ERCIM ACTIONS

FMICS 2005 — 10th International Workshop on Formal Methods for Industrial Critical Systems

by Mieke Massink and Tiziana Margaria

This 10th edition of the International Workshop on Formal Methods for Industrial Critical Systems (FMICS), a series of workshops organized by the homonymous ERCIM Working Group, was a good occasion to re-examine the use of formal methods in industry over the last ten years, and to outline a promising way forward for the next decade.

For ten years, the FMICS workshops Two invited speakers gave excellent pre- have striven to promote research on and sentations. Giorgios Koutsoukis, replac- support the improvement of formal ing Luis Andrades from ATX Software methods and tools for industrial critical SA, Lisbon, gave a presentation on the applications. They are intended to pro- experience of ATX with the application vide a common forum for scientists and of formal and rigorous techniques and Pedro Merino Gomez, University of industrial professionals to exchange methods in real projects. Christel Baier Malaga, Spain, giving a presentation of his experiences related to the development from the University of Bonn gave a pre- paper during the workshop. He received and application of formal methods. The sentation on the most recent develop- the 2005 EASST best-paper-award. Pedro FMICS Working Group has achieved ments in the quantitative analysis of dis- Merino Gomez has been elected the next both broad public visibility and good tributed randomized protocols. chair of the FMICS working group. interaction with the wider scientific community. These merits were recog- A special session was arranged for the nized by the ERCIM board of directors, presentation of a follow-up to the much The proceedings of the workshop were which granted the FMICS Working cited and widely discussed article ‘Ten published as ACM-SIGSOFT paper pro- Group the ERCIM award for the most Commandments of Formal Methods’ by ceedings and have also appeared in the successful Working Group of 2002. Jonathan P. Bowen and Michael G. ACM Digital Library. Selected papers Hinchey, which was published ten years will be invited for publication in a spe- Previous workshops were held in Oxford ago. Both authors joined the workshop to cial issue of the International Journal on (March 1996), Cesena (July 1997), present a perspective on ten years of the Software Tools for Technology Amsterdam (May 1998), Trento (July industrial application of formal methods, Transfer. 1999), Berlin (April 2000), Paris (July and set the stage for a lively discussion on 2000), Malaga (July 2002), Trondheim progress in the decade to come. The orga- The organizers wish to thank ESEC/FSE (July 2003) and Linz (September 2004). nizers are very happy that both authors for hosting the FMICS05 workshop and This year the FMICS workshop was co- chose the FMICS workshop as the forum taking care of many administrative located with the European Software in which to present their new ideas. aspects, ACM SIGSOFT for their spon- Engineering Conference (ESEC) and the sorship and ERCIM for its financial sup- ACM SIGSOFT Symposium on the The award for the best paper was granted port of the workshop. Additionally, the Foundations of Software Engineering this year to de la Camara, Gallardo, organizers would like to thank EASST (FSE). Together these comprise an inter- Merino and Sanan for their excellent (European Association of Software nationally renowned forum for paper on the application of model check- Science and Technology), ATX researchers, practitioners and educators ing to distributed software systems that Software, and the institutions CNR-ISTI in the field of software engineering, held use API Sockets and the network proto- and the University of Göttingen for sup- in the beautiful city of Lisbon in col stack for communications. The porting this event. Portugal. The workshop was organized award was granted with the support of by the Istituto di Scienze e Tecnologie the European Association of Software Link: della Informazione - A. Faedo of Pisa, Science and Technology (EASST). FMICS Working Group: Italy, and the University of Göttingen in Other papers presented at the FMICS05 http://www.inrialpes.fr/vasy/fmics/ Germany. Thirty participants from workshop included ‘An Approach to the academia and industry from about thir- Pervasive Formal Specification and Please contact: teen countries attended the workshop. Verification of an Automotive System’, Pedro Merino Gómez, ERCIM FMICS ‘Developing Critical Systems with PLD Working Group coordinator Fourteen contributions were selected Components’, ‘Test Coverage and Universidad de Málaga / SpaRCIM from 27 good-quality submissions, cover- System Development Based on Lustre’, Tel.: +34 952 132752 ing both industrially relevant theoretical ‘Integrated Formal Methods for E-mail: [email protected] URL: topics as well as industrial case studies. Intelligent Swarms’ and many others. http://www.lcc.uma.es/~pedro

6 ERCIM News No. 64, January 2006 JOINT ERCIM ACTIONS

Bits, Atoms and Genes Beyond the Horizon by Dimitris Plexousakis

The ‘Beyond the Horizon’ Project invites the research community in Europe to participate in defining research themes on Future and Emerging Information Society Technologies in the Seventh Framework Programme

In order to achieve the full potential of the ing to invent ICT components and sys- boundaries between information, mate- Information Society and to ensure the tems based on synthetic bio-molecular rial and life. In the coming decades, this long term competitiveness of Europe, vig- structures and beyond this, systems able cross fertilisation will intensify as scien- orous R&D activities in Information and to physically grow and self-repair. New tists from different disciplines learn from Communication Technologies (ICTs) and computing and communication schemes each other’s different ways of thinking. related disciplines are essential. This is are increasingly inspired by our bur- the key role of the Information Society geoning new understanding of the living Providing Input for FP7 Technologies (IST) programme within world. New simulation tools and systems Beyond-The-Horizon (B-T-H) is a coor- the forthcoming 7th Framework are being developed to model the living dination action funded by IST-FET and Programme. As part of IST, the coordinated by ERCIM. The pur- Future and Emerging Technologies pose of this project is to define (FET) activity has the role of stimu- Participate in the Europe-wide consultation pro- major challenges and promising lating the emergence and develop- cess that ERCIM is organising in close collabora- research directions that FET could ment of new IST-related disciplines tion with European S&T communities in order to support in the forthcoming FP7 in define the major challenges and promising and technologies, which promise to ICT-related strategic basic research directions that FET could support in the have significant scientific, indus- forthcoming Seventh Framework Programme. research areas. Through an exten- trial and societal impact. In this per- sive and systematic consultation of Visit http://www.beyond-the-horizon.net spective, FET is supporting long- the relevant science and technol- term, visionary, high-risk collabora- ogy communities in Europe, the tive basic research projects in project will deliver innovative, advanced strategic research areas related world from molecules to cells, organs, visionary and interdisciplinary research to ICTs. organisms and societies, and, for exam- directions and plans that would permit to ple, to predict effects of medicines or develop future ICT-related technologies Further advances in ICTs are increas- model the integration of artificial organs and their impact on society over the next ingly relying upon their synergy and with living organisms. 15 years. Six individual strategic cross-fertilisation with other scientific research fields as well as their cross- and technological fields. For example, This is particularly true between ICTs, links are the foci of the action, namely: we can see prospects in the miniaturisa- new materials, biology and the life sci- ‘Pervasive Computing and tion of ICT devices compatible and inter- ences, where progress in nano- and Communications (Thematic Group 1 or acting with living organisms, also allow- information technology is blurring the TG1)’, ‘Nanoelectronics and Nanotechnologies (TG2)’, ‘Security, Dependability and Trust (TG3)’, ‘Bio- ICT Synergies (TG4)’, ‘Intelligent and Cognitive Systems (TG5)’ and ‘Software Intensive Systems (TG6)’.

Throughout the year 2005, the six thematic groups have held brainstorming workshops for the purpose of identifying the emerging grand challenges in the respective areas. The workshops brought together eminent researchers from both academia and industry that embarked on drafting agendas for basic research in the Key speakers at the plenary workshop. From left: Gavriel Salvendy, six thematic areas. Summaries of the Purdue University, USA and Tsinghua University, China; Claude Cohen- findings of the groups are reported in Tannoudji, Ecole Normale Superieure, Paris, Winner of the 1997 Nobel short articles in this issue. More detailed Prize in Physics; Keith Jeffery, Director for IT of CCLRC and ERCIM reports are available through the pro- president; and Thierry Van der Pyl, Head of Unit, Future and Emerging ject’s web site. The reports will be final- Technologies, European Commission. ized in the next few months, after a wider

ERCIM News No. 64, January 2006 7 consultation with the relevant European Thematic Group 1: of people with these devices and the ser- research communities that will start soon. Pervasive Computing vices they offer (see Figure). Plenary Workshop Proposed Research Themes A Plenary Workshop in Paris, 12-13 and Communications In order to address the above, experts partic- December 2005 brought together repre- ipating in TG1 agreed to structure the final sentatives of the six aforementioned the- by Alois Ferscha research agenda along the following three matic groups for the purpose of examin- visionary challenges: ‘Networked Societies ing and consolidating the research chal- The vision impacting the evolution of of Artefacts’, ‘Evolve-able Systems’, and lenges and themes that were identified at Pervasive Computing and Communications ‘Human Computer Confluence’. the individual workshops and of explor- is that of an intuitive, unobtrusive and dis- ing new promising research challenges traction-free interaction with technology- The key technical problems and milestones that arise at the intersections of the dif- rich environments. In an attempt to bring that must be solved towards achieving this ferent areas. The debates were enriched interaction ‘back to the real world’ after an vision are summarised as follows: with varying perspectives by notable era of keyboard and screen interaction, com- invitees from the European research puters are being understood as secondary 1. Networked Societies of Artefacts community, including members of the artefacts, embedded and operating in the • deriving models for goal-oriented and IST Committee and representatives of background, whereas the set of all physical social behaviour the French Ministry for Research. Other objects present in the environment are • enhancing and enriching the communica- workshop special guests included understood as the primary artefacts, the tion fabric of societies, eg enabling oppor- Claude Cohen-Tannoudji, Winner of the ‘interface’. Instead of interacting with digi- tunistic networks 1997 Nobel Prize in Physics; Wendy tal information via traditional computing • harnessing dispersed content and manag- Hall, representative of the future means, Pervasive Computing aims at physi- ing information European Research Council; Anne • designing and developing space aware Cambon-Thomsen, member of the models. European Group on Ethics. Keith Jeffery, President of ERCIM; and, Ulf 2. Evolve-able Systems Dahlsten, IST Director, and Thierry Van • the development of viable, evolve-able der Pyl, IST-FET Head of Unit, systems (hardware, software, communi- Empowering people with a ‘pervasive European Commission. Prof. Gavriel computing landscape’, that is: cation fabric) - aware of their presence Salvendy (Purdue University, USA and - sensitive, adaptive and responsive • enabling the autonomic adaptation to Tsinghua University, China) shared an to the users’ capabilities, needs, unforeseen situations, interpreting con- habits and emotions ‘outsider’s view’ on the directions of - and ubiquitously, safely and text, and creating future aware behaviour securely accessible basic research in ICTs. The workshop - via natural interaction in support of both, long-term forward was very well-attended (80 delegates) evolution as well as short-term adaptivity and the discussion sessions were lively. and self-properties The complete programme, presentation Pervasive Computing – state of the art. • considering deterministic, nondeterminis- slides and photographic material are tic and bio-inspired paradigms, including available through the project web site. cal interaction with digital information, ie, stochastic approximation, interaction by manipulating physical arte- • coping with the fundamental issue of B-T-H is now entering its second stage facts via ‘graspable’ interfaces. It links the scale. during which the Thematic Group ‘atoms of the physical world’ with the ‘bits reports are disseminated to the European of the digital world’ in such a way, that 3. Human Computer Confluence research community at large for feed- every physical artefact is considered as • supporting invisible, implicit, natural, back and consultation. The reports are being both representation of and control for embodied, and even implanted interac- accessible through the action’s web site. digital information. tion, Research communities in Europe are • considering qualitative aspects such as invited to provide their comments The challenges of Pervasive Computing are user experience and user behaviour. through the on-line consultation mecha- dominated by the ubiquity of a vast mani- nism on the B-T-H website. fold of heterogeneous, small, embedded and Participate in the online consultation of this possibly mobile artefacts, the evolvability of report from 1 February to 31 March 2006 at Please contact: their population, functionalities and interop- http://www.beyond-the-horizon.net Dimitris Plexousakis, ICS-FORTH eration, the ability of perceiving and inter- B-T-H Scientific Coordinator preting their situation, the autonomicity of TG1 Coordinator: Alois Ferscha E-mail: [email protected] their goal-oriented behaviour, the dynamic- University of Linz, Austria ity and context-adaptation of services E-mail: [email protected] Jessica Michel, ERCIM Office offered, the ad-hoc interoperability of ser- B-T-H Administrative and Financial Coordinator vices and the different modes of interaction E-mail: [email protected]

8 ERCIM News No. 64, January 2006 JOINT ERCIM ACTIONS

Thematic Group 2: tives such as high performance, reliabil- applications to microwave signal pro- ity, correctness, ease of programming, to cessing, mechanically detecting mag- Nanoelectronics and developments of advanced, potentially netic resonance imaging, bio-sensors interfering or unreliable nanoscale- and bio-actuators, micro-nanofluidics, Nanotechnologies devices. An important element is to build single molecule sensing and-analysing, suitable simulation methods and design data storage and operation at the quan- by Colin Lambert tools that link nano-scale device models tum limit. with higher-level heterogeneous system Experts participating in this thematic design environments 6. Nanotechnologies for quantum- group recommend sustained investment coherent systems, aimed at exploiting in research to underpin the ‘More of 2. Interfacing nano-scale biology with quantum effects in solid state devices. It Moore’, ‘More than Moore’ and nano-electronics, including bio-nano includes investigating new types of ‘Beyond Moore’ technology drivers: transduction and growable electronics. solid-state qubits and scalable coherent • The ‘More of Moore’ approach is This would provide the basic hardware systems to build large-scale coherent focused on delivering the structures to develop research objectives systems and practical quantum comput- ‘International Technology Roadmap under B-T-H TG5 (Intelligent and ers, and at addressing the ambitious for Semiconductors’ (ITRS) for late Cognitive Systems) and could lead to cir- materials science challenges associated complementary metal oxide semicon- cuits and connections which grow, shrink with the engineering of solid-state ductor (CMOS) and post-CMOS sys- or reconfigure according to demands on qubits and quantum coherent systems in tems, and in particular, how to con- functionality, thereby impinging on solid-state environments. tinue Moore’s law beyond the pre- activities in TG2 involving evolvable dicted 22 nm node in 2011. hardware and emergent design. Participate in the online consultation of this • The ‘More than Moore’ approach is report from 1 February to 31 March 2006 at focused on delivering greater func- 3. Future interconnects for heteroge- http://www.beyond-the-horizon.net tionality through heterogeneous inte- neous system integration. Objectives gration of nanosystems with elec- include higher integration density, less TG2 Coordinator: Colin Lambert tronic, optical, magnetic, chemical, I/Os, shorter wires, lower power dissipa- University of Lancaster, U.K. biological, mechanical and other function and higher speed. Promising E-mail: c [email protected] tions: Research themes which extend research directions include the use of ‘Beyond Moore’ in terms of their nanotubes, nanowires, recently-synthe- potential to deliver disruptive tech- sised molecules, nonlinear wave propa- nologies include new paradigms such gation and 3d architectures Another Thematic Group 3: as intra-molecular computing and direction is to emulate natural systems solid-state quantum information pro- such as nerve bundles, which show an Security, Dependability cessing (SSQIP), addressing the example of unidirectional, self-restored emerging field of engineering coher- signal propagation, chemically assisted and Trust ent solid-state quantum systems. guided growth and life-long repair capability. To avoid problems associated by Michel Riguidel Future disruptive technologies and with high-density interconnects, non- breakthroughs are likely to come from local processing in non-charge-based ICT security balances freedom and the progress in a range of rapidly-develop- devices and interconnect-lean architec- desire to protect tangible and intangible ing areas The medium-term impact of tures such as cellular automata could assets, ensures the immunity of applica- many of these technologies may initially also be explored. tions and system resilience, and instils occur in niche ‘More than Moore’ areas, and preserves confidence in digital, criti- which would provide economic benefits 4. Post-CMOS memory, storage and cal infrastructures. At the smallest level, and stimuli before impacting on the logic, aimed at identifying nano-devices nanotechnology, quantum communica- ‘More of Moore’ or “Beyond Moore” that integrate gracefully with CMOS tion and cryptography offer new oppor- challenges. and architectures that exploit the advan- tunities to tackle ICT security. tages of both CMOS ‘hosts’ and these Proposed Research Themes new nanotech blocks. The research Embedded sensors and devices can form The following promising research would investigate a range of information ad-hoc networks requiring new mecha- themes were identified by experts of this carriers such as electrons, spins, pho- nisms for establishing trust when sharing Thematic Group: tons, phonons, atoms, molecules, information or resources. New mechanical state and material phase. paradigms come to the foreground, such 1. ‘System-ability’of emerging ICT tech- as service architectures that compose nologies and devices, involving multi- 5. Nanoelectromechanical systems services from lower level modules, peer- disciplinary teams of system architects (NEMS), including VLSI-like arrays of to- peer systems characterized by their and nano-technology researchers. This sensors, probes and sources and nano- remarkable robustness and resilience research would link system-level objec- object-based NEMS with potential against attack, and biological defence

ERCIM News No. 64, January 2006 9 mechanisms which may inspire new 3. Quantum Technology Thematic Group 4: breakthrough technologies. At a larger and Cryptography scale, the completion of the Galileo Nature can provide us with resources to Bio-ICT Synergies satellite navigation system around 2009 secure our information and communica- will create ever more sophisticated pos- tion systems. The possibility provided by by Fernando Martin-Sanchez sibilities for positioning with implica- Quantum technology to offer secret bits tions for both security and privacy. of information among authenticated dis- The prospect of having ICTs that would tant partners, as well as truly random val- be more life-like and thus for instance Proposed Research Themes ues, could serve as building blocks of more robust, self-regulating, adaptive, In view of the above, the following many protection mechanisms. Quantum evolving, environment friendly and easy emerging research themes were identi- technology and quantum computing to live with, has long stimulated com- fied by experts participating in this might also represent a major threat for puter scientists to look for inspiration in Thematic Group: current cryptographic algorithms and the biological domain. Most bio-inspired mechanisms. We should also study the ICT techniques and methods were 1. Ambient Trustworthiness assumptions under which Quantum designed on the basis of biological The mass diffusion of digital systems Computers (QC) may act and their con- knowledge available at the '80s. In the must be enabled with built-in mecha- sequences on current and future crypto- meantime, biology has brought a wealth nisms for enhancing trust and confidence graphic methods, as well as the develop- of new knowledge – for instance on on their usage. Common security mecha- ment of new QC resistant cryptographic development, action, perception, home- nisms mainly based on boundaries and techniques. ostasis and learning – that can now be firewall protection mechanisms do not exploited to fuel further advances in ICT scale with respect to new complex sys- 4. Assessability and Verifiability systems. Conversely, efficient algo- tems. We should imagine different Assessing and proving the trustworthi- rithms can be used to reflect back to biol- mechanisms such as the ones inspired by ness of a complex system is a major ogy, for instance to enable more efficient the living world: immune and self-heal- issue. During the last years many tech- drug design, to move toward personal- ing systems. We should consider auto- niques have been developed, especially ized medicine, or to distill a more syn- nomic, evolvable and adaptive security in the dependability community. Yet, the thetic view of the biological field. mechanisms, which will require new scale of new ICT systems and the kind of Beyond such cross-fertilization, how- semantic models managing the complex- threats and assumptions about their oper- ever, maturing technologies and know- ity of ambient intelligence environments ational environment pose new chal- ledge from both sides are paving the way where humans and devices may jointly lenges. Different metrics, modelling for a deeper convergence between the function and interact. Security systems tools and observation mechanisms are biological and the ICT domain which and cryptographic mechanisms must be needed. The capability of measuring the will lead to systems with tightly inte- scaled down for inclusion in small tolerance to attacks is crucial in new sys- grated biological and technological com- devices (even at nano-scale) with spe- tems that due to their logical and physi- ponents and to radical innovations, for cific requirements for energy consump- cal diffusion are susceptible to ‘attack’. instance in computer architectures, inter- tion and computation power. There is a need to develop a discipline of faces, prosthesis and implants. software and system security based on 2. Dynamicity of Trust high-level verifiably secure program- The aim of this thematic group is to iden- Lack of trust either on the cyber-infra- ming. This calls for an approach based tify a long-term research agenda that structure (due to frequent attacks) or the on efficiently verifiable mathematical would unlock the full potential of Bio- difficulties to model trust relationshipsa- proofs showing compliance to policies ICT synergies over the next few decades. mong different entities (human and digi- (expressing safety, security, or function- How can we meaningfully combine bio- tal ones) is one of the main barriers for ality constraints). logical and technological perspectives the establishment of a true Information and exploit what is learned to advance Society. In future ICT systems with bil- Participate in the online consultation of this both biology and technology, effectively lions of interconnected devices, the report from 1 February to 31 March 2006 at linking the realms of information, mate- capability of managing and negotiating http://www.beyond-the-horizon.net rial and life? The vision of technology trust relationships that foster cooperation becoming like a ‘second nature’ would is crucial. The understanding of how TG3 Coordinator: Michel Riguidel have tremendous social, economic and trust emerges and evolves, as well as of ENST-Paris, France industrial impact. Not only would it lead the related notions of reputation forma- E-mail: [email protected] to new types of computational hard- and tion, monitoring and evolution are software; it would also improve, among mandatory. Security-based trust and others, manufacturing, medicine, the trust-based security are two emerging energy sector and the quality of the envi- areas of interest. A deeper understanding ronment and of life in general. This will of trust needs the involvement of require radical interdisciplinarity, research expertise from several fields including the whole range of life sci- such as economy and sociology. ences, but also the nano- and neuro-sci-

10 ERCIM News No. 64, January 2006 JOINT ERCIM ACTIONS

Connecting cial limbs). In this sense, main chal- the natural, lenges include: (a) developing new digital and information theories and modelling tech- artificial niques to capture how biological systems worlds: cell realise basic capabilities of living sys- modelling, tems at different granularities, (b) devel- synthesis and oping ways to validate such theories with hybrid respect to real biological systems, (c) systems. interfacing between the living and the artificial world, including restoration or substitution of human capabilities (eg, using implants), (d) providing sensor/motor interface to the body, (e) extending the range of capacities beyond ences. This being said, this group ple devices operating in a highly parallel perception and action, including for emphasises the ethical dimension of the fashion at comparatively low speed and instance memory, resistance to bacteria vision, as well as the well known but with very low power dissipation. and viruses, or interfacing directly in unresolved difficulty of genuine multi- metabolic processes (‘cyber-drugs’). disciplinary research as key extra-scien- 2. Bio-Inspired Strategies of Growth, tific hurdles to be addressed. Adaptation and Evolution Participate in the online consultation of this In contrast to technological systems, bio- report from 1 February to 31 March 2006 at Proposed Research Themes logical systems have extraordinary capa- http://www.beyond-the-horizon.net Three main research themes were identi- bility to change over time. They grow, fied as pillars for exploiting the Bio-ICT adapt, self-assemble, replicate, heal, self- TG4 Coordinator: Fernando Martin-Sanchez synergies potential, namely ‘New organise, evolve. The second theme thus Institute of Health Carlos III, Madrid Spain Computational Modelling Paradigms’, concerns studying how technological E-mail: [email protected] ‘Bio-Inspired Strategies of Growth, systems can grow, change, adapt, organ- Adaptation and Evolution’ and ‘Bio-ICT ise, heal and evolve to match, over long Artefacts’ (see Figure). periods of time, evolving needs whilst being compatible with natural processes Thematic Group 5: 1. New Computational Modelling of change that surround them, for Paradigms instance when dispersed in the environ- Intelligent and A major research challenge is the for- ment, or when implanted. These different malization and development of computa- strategies of change are not independent Cognitive Systems tional paradigms that are capable of cap- but operate at different time scales and turing, relating and integrating the differ- either at the individual or population by Rolf Pfeifer and Alois Knoll ent levels of complexity in biological level. We propose an interdisciplinary systems - from the molecule to the cell, exploration of adaptation, learning, self- The grand challenge is that of building tissue, organ, system, individual, popula- organisation, evolution and other emer- intelligent and cognitive systems that are tion and ecosystem. Some of the techni- gent functionalities of living systems for capable of acting in the real world and cal challenges involved are: (a) connect- the design of new computing models, interacting with humans in natural ways. ing models at different levels of abstrac- algorithms and software programming tion, (b)connecting discrete with contin- paradigms. Moreover, there is enormous A promising research area tackling this uous models, (c) dealing with inconsis- potential for applying this at the physical challenge is the development of so- tent, competing models and (d) fitting level, in particular at the nano- and micro called complete agents, that is, agents models to data (data driven modelling). scale, to develop new types of growing that are embodied and self-sufficient (ie, Other than contributing to systems biol- and evolving hardware (eg, memory or they can sustain themselves over ogy, this research would lead to new computing capacity grows when needed) extended periods of time), situated (ie, computational architectures inspired by and intelligent materials that could be they can acquire information about the natural information processing structures applied in a variety of ambient interfaces. environment through their own sensory eg in the brain or molecular processing systems), and autonomous (ie, they func- within the cell, but applicable in different 3. Bio-ICT artefacts tion independently of external control). domains than the biological ones. The third theme of research in modelling For ‘intelligent and cognitive systems’, both the organizational and phenomeno- participants in this thematic group rec- A thorough understanding of information logical features of living systems is to ommend to focus research efforts within processing in biological systems would seamlessly integrate artificial entities the framework of embodiment. In this lead to new computing systems based on into biological systems and processes framework, intelligence and cognition ‘wet-ware’ or other life-like hardware, (classical examples include artificial are properties that emerge as an agent consisting of very large numbers of sim- retinas or physiologically coupled artifi- interacts with its environment.

ERCIM News No. 64, January 2006 11 Proposed Research Themes erogeneous systems, embedded systems In order to meet the above long-term for automotive and avionics applica- challenge, the following research themes tions, telecommunications, wireless ad need to be pursued: hoc systems, business applications with an emphasis on web services etc. 1. Mind-Body Co-Development and Co-Evolution Our daily activities increasingly depend In order to maximally exploit the design on complex software-intensive systems power of evolution and development, that are becoming ever more dis- controllers and robot morphologies have Task Group 5 workshop participants tributed, heterogeneous, decentralized to evolve simultaneously. The permanent Giulio Sandini and Olaf Spoorns learn and inter-dependent, and that are oper- interaction of the body of an agent with about the artificial whisker system ating more and more in dynamic and the environment during growth enables developed by Miriam Fend of the Artificial often unpredictable environments. its ‘mind’ to develop. We need to under- Intelligence Laboratory in Zurich in the These trends will continue to amplify stand and model developmental pro- IST-FET-funded Artificial Mouse Project in the future, while requirements for cesses in nature and build embodied arti- (AMouse, http://www.amouse.de/). quality of service, security, and trust ficial systems that are inspired by them; will dramatically increase. Current embed them into evolutionary processes, interaction of an agent with the environ- engineering methods and tools are not and study interaction of physical and ment, behaviour is emergent, meaning powerful enough to design, build, information processes during this devel- that it cannot be understood (and deploy, and maintain such systems. opment. These also require physical designed) on the basis of the internal con- Moreover, to continue to work depend- growth (see the following point). trol program (or ‘brain’) only. The ques- ably with them, such systems would tion then is: how can we design purposive have to exhibit adaptive and even antic- 2. Systems and Materials that can (goal-directed) agents without destroying ipatory behaviour. Today's grand chal- Physically Grow the emergent nature of their behaviour? lenge is to develop practically useful Recent research strongly suggests that and theoretically well-founded princi- physical instantiation and materials play In each of these research areas real phys- ples, methods, algorithms and tools for an essential role in system behaviour. ical embodiment plays an essential role. programming and engineering such Also, through growth, biological organ- Simulation work (‘embodied agent sim- future software intensive systems isms can form highly complex morpho- ulations’) could form an important part throughout their whole life-cycle. logical structures. In this respect there are of the endeavour. promising starting points (eg, modular Proposed Research Themes robotics, self-assembling materials, etc). Participate in the online consultation of this Among the many promising areas for Although it is not expected to have grow- report from 1 February to 31 March 2006 at future research, the participants in this able materials that match biological http://www.beyond-the-horizon.net Thematic Group have identified three capacities available any time soon, some crucial areas: Engineering adaptive soft- aspects of physical growth can be studied TG5 Coordinator: Rolf Pfeifer ware-intensive systems; managing through advanced principles of autono- University of Zurich, Switzerland diversity in knowledge; and eternal soft- mous modular systems that optimally E-mail: [email protected] ware-intensive systems. adapt to a task and the environment. 1. Engineering Adaptive 3. Morphological Computation Software-Intensive Systems Shape and the materials (eg, muscles) per- Thematic Group 6: The current approach, where systems are form important functions for an agent in mainly assembled at design time does not real time, an idea that bypasses the con- Software-Intensive scale to pervasive, highly dynamic sys- cept of classical Turing computation tems. The emergent behaviour of systems architecture. Morphological computation Systems is an unavoidable fact that must be here refers to processes based on shape exploited during the system's life time, in (eg, molecules/DNA, modules of a modu- by Martin Wirsing order to scale to the level of complexity lar robot, shape of a limb) and material we are witnessing. Systems will no longer properties (eg of the muscle-tendon sys- Software has become a central part of a be produced ab initio, but more and more tem). The challenge will be to explicitly rapidly growing range of applications, as compositions and/or adaptations of apply morphological concepts in theoreti- products and services from all sectors of other, existing systems, often performed cal and practical explorations of embod- economic activity. Systems in which at runtime as a result of a process of evo- ied artificial systems. software interacts with other software, lution. The challenge is to develop algo- systems, devices, sensors and with peo- rithms, methods, tools and theoretical 4. Design for Emergence ple are called software-intensive sys- foundations that enable effective design As behaviour is always the result of the tems. Examples include large-scale het-

12 ERCIM News No. 64, January 2006 JOINT ERCIM ACTIONS

Quantum Information Processing and Communication by harnessing, controlling and using the Preparation for FP7 in Future and Emerging Technologies Unit, DG INFSO, EC effects of emergent system properties. In view of the forthcoming FP7, the Unit Future and Emerging Technologies (FET) of 2. Managing Diversity in Knowledge DG Information Society and Media is now carrying out a Europe-wide consultation We are facing an unforeseen growth of process with the relevant S&T communities in a number of ICT fields. The objective the volume and complexity of the data, is to define the major challenges and promising research directions that FET could content and knowledge being produced. support in FP7 (see B-T-H articles in this issue). In knowledge management and engineering, the ‘usual’ approach is to take Quantum Information Processing and Communication (QIPC) is now a well established into account, at design time, the possible scientific field which opens unconventional perspectives for information processing. It future dynamics, most commonly by exploits fundamentally new modes of computation and communication with the aim to designing a global reference representa- understand the quantum nature of information and to learn how to formulate, manipulate, tion schema and by codifying into it all and process it using physical systems that operate on quantum mechanical principles the possible diverse knowledge compo- (control of coherent superpositions of quantum degrees of freedom – qubits). Today, there nents. As applications become more and is a significant world-wide effort to advance research in QIPC, which has led to a deeper more open and complex, this top-down and broader understanding of information theory, of computer science and of the approach shows its limits. We need a fundamental laws of the quantum world. Advances in QIPC could soon lead to new new, bottom-up, approach for managing technologies and devices that hold the promise to radically change the way we compute knowledge where the different know- and communicate. ledge parts are designed and kept ‘locally’ and independently and where Since its early steps, European scientists have been at the forefront of QIPC research. So far, new knowledge is obtained by the FET is the sole part of the IST Programme that has been supporting QIPC research and has design- or run-time adaptation of such been very successful in attracting the best research teams in Europe. While the field has now different knowledge parts. This calls for reached a certain degree of maturity and there is critical mass in Europe in the main sub- developing adaptive or even self-adap- fields, it is still necessary to further expand and strengthen activities at the European level. In tive knowledge systems that are able to view of the forthcoming FP7, FET and the European QIPC research community have actively manage diversity in knowledge by har- been working during the last year towards the elaboration of a common European research nessing, controlling and using the effects strategy in the field. These efforts have culminated in the publication of a technology roadmap of emergent knowledge properties. on ‘Quantum Information Processing and Communication: Strategic report on current status, visions and goals for research in Europe’. This report was written by the most prominent 3. Eternal Software-Intensive Systems scientists in the field and, after a wide consultation within the QIPC research community, it is Information, and the tools to work with now published on the FET web site (http://www.cordis.lu/ist/fet/qipc.htm#prepfp7). it, represent one of society's most important assets. From a cultural as well as an The QIPC roadmap presents in a comprehensive way the state-of-the-art, the medium and economic point of view, it is essential to long term goals and the visions and challenges for the future. It includes an overview of enable continuous and up-to-date access FET activities, a description of national research programmes and the worldwide research to long-lived and trustworthy informa- position of Europe in QIPC. The main bulk of the document is devoted to a scientific tion systems, as well as to guarantee that assessment of current results and an outlook of future efforts. It covers three main research the corresponding information systems directions: quantum communication, quantum computing and quantum information science, do not age and break but are able to pre- as well as the interactions and interdependences between them. The document stipulates serve and update their original function- the need for further support in these three research directions, as well as to keep a diversity ality and properties in a machine inde- of experimental realizations and to look for synergies between them in order to reach pendent way by re-programming them- concrete objectives. Integration across different disciplines and between different selves to take into account new contexts. experimental approaches is considered crucial for the further advancement of QIPC in In other terms, the challenge is to orga- Europe. Prospects for applications and commercial exploitation are equally discussed. The nize software-intensive systems so that roadmap is a living document, which will be periodically updated in order to serve as a they can survive and evolve in a con- guideline both to scientists and decision makers. stantly changing world. In parallel to the strategic report, the research community in collaboration with FET has Participate in the online consultation of this produced the publication “QIPC in Europe”. It is a collection of 30 articles in “Scientific report from 1 February to 31 March 2006 at American” style written by 58 of the most prominent experts in Europe. It gives a balanced http://www.beyond-the-horizon.net overview of QIPC research in Europe and refers to work accomplished within FET and nationally funded projects. These two documents complement each other and are important TG6 Coordinator: Martin Wirsing milestones along the way towards elaborating a common European strategy in QIPC. They Universitat Munchen, Germany are both published in the FET QIPC proactive initiative web site and by the Publications E-mail: Office of the European Commission. [email protected] Link: http://www.cordis.lu/ist/fet/qipc.htm

ERCIM News No. 64, January 2006 13 News from W3C

The Web on the Move

Successfully attended with over 150 participants, the first out- Mobile Web Initiative sponsors were present on two panels: reach public Mobile Web Initiative (MWI) event, held in ‘Mobile Web Challenges and Potential’ and ‘Putting the Vision London on 15 Novenber 2005, focused on current mobile Web into Practice’, chaired respectively by Rotan Hanharan, chief challenges and opportunities. The program opened with a video architect at MobileAware, and Daniel Appelquist, senior tech- welcome from Tim Berners-Lee, followed by a presentation on nology strategist for the Vodafone Group. All presentations are the vision and ambitions for the mobile Web by Philipp linked from the agenda available on the event home page. Hoschka, Deputy Director for W3C Europe and W3C’s Mobile Web Initiative leader. The mobile Web goes where the users go. Instead of running back to their computer for Web access, users will now have Web access when and where they need it, using their mobile device. To that effect, Mobile Web Initiative (MWI) participants are building a database of device descriptions and developing best practices for the creation of mobile friendly Web sites.

MWI is led by key players in the mobile production chain, including authoring tool vendors, content providers, handset manufacturers, adaptation providers, browser vendors and mobile operators. Current MWI sponsors are: Ericsson, France Telecom, HP, Nokia, NTT DoCoMo, TIM Italia, Vodafone, Afilias, Bango.net, Drutt Corporation, Jataayu Software, Mobileaware, Opera Software, Segala M Test, Sevenval AG, Rulespace and Volantis.

Links: MWI UK event: http://www.w3.org/2005/11/mwi-ukevent.html Mobile Web Initiative (MWI) event in London. W3C Mobile Web Initiative: http://www.w3.org/Mobile/

W3C to Internationalize While these attributes are critical, additional attributes may be even more important to specific languages. For example, and Secure Voice Browsing Mandarin Chinese, the most widely spoken language in the world today, also has the notion of tones - the same written Following the successful technical Workshop held in Beijing character can have multiple pronunciations and meanings last November 2005, and taking into account valuable inputs based on the tone used. Given the profusion of cellphones in from the VoiceXML Forum, the W3C announced new work on China - some estimate as high as over one billion - the case for extensions to components of the Speech Interface Framework extending SSML for Mandarin is clear in terms of sheer market which will both extend Speech Synthesis Markup Language forces. Including extensions for Japanese, Korean and other functionality to Asian and other languages, and include speaker languages will ensure that a fuller participation possible of the verification features into the next version of VoiceXML, ver- world on the Web. sion 3.0. Addressing both areas expands both the reach and functionality of the framework. Users of telephony services and the Web are also demanding speaker verification. Identity theft, fraud, phishing, terrorism, The Speech Synthesis Markup Language (SSML), a W3C and even the high cost of resetting passwords have heightened Recommendation since 2004, is designed to provide a rich, interest in deploying biometric security for all communication XML-based markup language for assisting the generation of channels, including the telephone. Speaker verification and synthetic speech in Web and other applications. The essential identification is not only the best biometric for securing tele- role of the markup language is to provide authors of synthesiz- phone transactions and communications, it can work seam- able content a standard way to control aspects of speech such as lessly with speech recognition and speech synthesis in pronunciation, volume, pitch, rate, etc. across different synthe- VoiceXML deployments. sis-capable platforms. Link: Voice Browser Activity: http://www.w3.org/Voice/

14 ERCIM News No. 64, January 2006 Call for Participation W3C Continues ICANN Participation W3C Workshop: W3C is pleased to announce the nomination of Daniel ‘Toward a More Secure Web’ Dardailler, W3C Associate Chair based in Europe, as W3C liaison to the Internet Corporation for Assigned Names and New York City, USA, 15-16 March 2006 Numbers (ICANN) Board of Directors. Thomas Roessler will serve on the 2006 ICANN Nominating Committee (NomCom). Web security today critically depends on Transport Layer W3C’s involvement in ICANN and in the post-WSIS forum, Security (TLS), an IETF protocol that is wrapped around HTTP soon to be launched, should help preserve the operational stabil- transactions to provide endpoint authentication and communi- ity of the Internet and the Web in a transparent and open way cations privacy. Ongoing ‘phishing’ attacks demonstrate that while ensuring its unfragmented growth based on contributions these security measures fail in practice: while the currently from the international community. available mechanisms are technically solid, implementations often don’t succeed in making users aware what kind of secu- Call for Participation rity is actually in place, and with whom they are actually communicating. As a result, attackers can bypass these security Workshop on the Ubiquitous Web mechanisms without users noticing. Position papers are due 10 February for the W3C Workshop on This Workshop aims to concretely identify a range of issues the Ubiquitous Web to be held 9-10 March 2006, hosted by faced by those who wish to use the Web as a secure environ- Keio University in Tokyo, Japan. The ‘Ubiquitous Web’ seeks ment for tasks ranging from basic browsing to the most special- to fulfill the potential of the Web for distributed applications ized application. In particular, the participants will look at ways that adapt to the user’s needs, device capabilities and environ- to help address the current threats on the Web that are caused mental conditions. by the present lack of comprehensible and transparent Web authentication. The Workshop is expected to focus on near- Link: http://www.w3.org/2005/10/ubiweb-workshop-cfp.html term improvements that can be realized in browsers and through best practices coordinated between browser vendors and e-commerce service providers. Experiences and use cases W3C Web Services Addressing from the financial services industry are expected to inform the discussion. Interoperability Event

The Call for Participation solicits position papers from Web The W3C Web Services Addressing Working Group held an security experts, software developers, browser manufacturers Interoperability event on 17-18 January in Vancouver, BC, and their customers regarding usability and transparency of Canada. Participants tested the Web Services Addressing Web authentication. The goal is to identify methods to make family of W3C specifications. The group has invited interested secure, trustworthy browsing easy. parties who have implemented Web Services Addressing 1.0: Core, SOAP Binding and/or WSDL Binding. The Workshop is chaired by Daniel Schutzer (Citigroup), and Thomas Roessler (W3C). Position papers can be submitted by Link: W3C Web Services Activity: http://www.w3.org/2002/ws/ email until 25 January 2006. The Workshop takes place in New York City, USA, on 15 and 16 March 2006, and is hosted by Citigroup. W3C Seminar on Web Services

Links: Paris, 6 March 2006 Workshop: http://www.w3.org/2005/Security/usability-ws/ Call: http://www.w3.org/2005/Security/usability-ws/cfp.html W3C is organizing a morning of presentations related to the use of Web Services in the industry today. The program of the half day will feature presentations from W3C Members. Web Services are software systems that support interoperable Latest W3C Recommendation machine-to-machine interactions over a network. They provide an universal glue between different applications within the • Synchronized Multimedia Integration Language (SMIL 2.1) enterprise, or between different companies. This seminar is 13 December 2005, Dick Bulterman, CWI, Guido Grassel, funded by the Web Services and Semantics (WS2) project, Nokia, Jack Jansen, CWI, Antti Koivisto, Nokia, Nabil financed by the European Commission’s FP6 IST Programme. Layaïda, INRIA, Thierry Michel, W3C, Sjoerd Mullender, Attendance to the seminar is free and open to the public. CWI, Daniel Zucker, Access Co., Ltd. Links: A complete list of all W3C Technical Reports: WS2 seminar: http://www.w3.org/2006/03/ws2-seminar.html http://www.w3.org/TR/ WS2 project: http://www.w3.org/2004/WS2/

ERCIM News No. 64, January 2006 15 EUROPEAN SCENE: OPEN ACCESS

Open Access: An Introduction

by Keith G Jeffery

Open Access (OA) means that electronic scholarly articles are available freely at the point of use. The subject has been discussed for over 10 years, but has reached a crescendo of discussion over the last few years with various declarations in favour of OA from groups of researchers or their representatives. The UK House of Commons Science and Technology Committee considered the issue in 2004, reporting in the summer in favour of OA. This indicates the result that libraries with restricted budgets (ie all of them!) are importance of the issue, and led to statements from large no longer providing many journals needed by researchers. research funding bodies such as the Welcome Trust and the Research Councils UK. Just reward: There is also concern that in traditional scholarly publishing, most of the work (authoring, reviewing, editing) is Motivations done freely by the community and that the publishers make exces- Ethics: There is an ethical argument that research funded by the sive profits from the actual publishing (making available) process. public should be available to the public. Since research is an In conventional publishing, the institution subscribes to the publi- international activity, this crosses national boundaries. cation channel to obtain electronic access or paper copies.

Research Impact: The Internet provides an opportunity. Modern Types of Open Access harvesting techniques and search engines make it possible to dis- At this stage it is important to distinguish several dimensions of cover publications of relevance if they are deposited in an OA the issue: OA can be delivered in two ways: repository with a particular metadata standard. If all authors did • ‘green’: the author can self-archive at the time of submission this then the world of research would be available ‘at the finger- of the publication (the ‘green’ route) whether the publication tips’. There is evidence that articles available in an OA repository is grey literature (usually internal non-peer-reviewed), a have more accesses (readers), citations and therefore impact. peer-reviewed journal publication, a peer-reviewed conference proceedings paper or a monograph Costs: There is concern over the hindrance to research caused • ‘gold’: the author or author institution can pay a fee to the pub- by the cost of journal subscriptions, whether electronic or lisher at publication time, the publisher thereafter making the paper. These costs run well above the rate of inflation with the material available ‘free’ at the point of access (the ‘gold’ route). The two are not, of course, incompatible and can co-exist.

The ‘green’ route makes publications available freely in parallel with any publication system but is not, itself, publishing. The ‘gold’ route is one example of electronic publishing. At present it is much more common to have non-OA electronic access to publications in a publisher’s database for a subscription fee.

The second dimension to be distinguished is the timing and quality aspect: preprints are pre-peer-review articles, postprints are post-peer-review and post-publication articles while eprints can be either but in electronic form.

16 A third dimension is white/grey literature. White literature is Green Route The author can self-archive at the time of submission of the publication whether the peer-reviewed, published articles while grey is preprints or inter- publication is grey literature, a peer-reviewed nal ‘know-how’ material. Of course there are usually many inter- journal publication, a peer-reviewed conference proceedings paper or a monograph esting relationships between grey and white articles (see Table). Golden Route The author or author institution can pay a fee to the publisher at publication time, the publisher thereafter making the material available ‘free’ at Barriers to Open Access the point of access. Loss of publisher income: The major objection to ‘green’ self- Preprints Preprints are articles that are pre-peer-review archiving comes from publishers and learned societies (many Postprints Postprints are articles that are post-peer-review eprints eprints can be either preprints or postprints but in of which depend on subscriptions to their publications) who electronic form fear that ‘green’ OA threatens their business viability. To date White Literature White literature is peer-reviewed, published articles there is no evidence that ‘green’ archiving harms the business Grey Literature Grey literature is preprints or internal ‘know-how’ material model of publishing. There is evidence that ‘green’ archiving increases utilisation, citation and impact of a publication. Dimensions of Open Access publishing. Whilst the major commercial publishers provide additional value-added services that could offset the impact of OA on current business models, the impact on learned societies may require new business models to be developed. the case of ‘green’ open access the OAI-PMH (Open Access Initiative – Protocol for Metadata Harvesting) facility links Copyright: Copyright agreements between authors and pub- OA repositories so that all repositories obeying the protocol lishers may inhibit the ‘green’ route. However, to date, between can be harvested and their contents are available freely. 80 and 90% of publication channels (the variability depends on exactly what is counted) allow ‘green’ author deposit although A Word of Warning some insist on an embargo period before the publication is Digitally-created articles rely heavily on both the metadata available for OA. In contrast some publishers of journals – of record and the articles themselves being deposited. International which ‘Nature’ is the most well-known – do not demand copy- metadata standards and protocols must be applied to repositories right from the author but merely a licence to publish, leaving so that harvesting across repositories can take place. To ensure copyright with the author or their institution. that research output material is available for future generations, curation and preservation issues must be addressed. Green Open Access Repositories There are two kinds of ‘green’ OA repository: Speculation: Future • Thematic: where authors deposit in a (usually) central repos- Looking to the future speculatively, it is possible to imagine itory used by the community and maintained by an appropri- ‘green’ OA repositories becoming commonplace and used heav- ate institution and where relevant material on a subject area is ily. At that point, some argue, one could change the business collected together. The best known example is arXiv model so that an author deposits in an open access ‘green’ repos- • Institutional: where the authors deposit in a repository main- itory but instead of submitting in parallel to a journal or confer- tained by their institution thus collecting together in one ence peer-review process, the peer-review is done either by: place the research output of that institution. This has the • a learned society managing a ‘college’ of experts and the advantage of responsibility of ownership and some possible reviewing process – for a fee paid by the institution of the management control/encouragement of deposit. author or the author; • allowing annotation by any reader (with digital signature to There are available open source systems for ‘green’ reposito- ensure identification/authentication). ries; the best known being ePrints, DSpace, Fedora and ePubs. The former peer-review mechanism would maintain learned Advantages of Open Access societies in business, would still cost the institution of the author The major advantage of OA is research impact – the available or the author but would probably be less expensive than pub- e-article is likely to have more accesses, citations and impact. lisher subscriptions or ‘gold’ (author or author institution pays) However, there are additional advantages: open access. The latter is much more adventurous and in the • Links: Electronic availability of a publication (whether spirit of the internet; in a charming way it somehow recaptures ‘green’ or ‘gold’) has another advantage; it is possible to the scholarly process of two centuries ago (initial draft, open crosslink the publication to any research datasets and software discussion, revision and publication) in a modern world context. used in producing the paper; this improves ‘the research pro- It is this possible future that is feared by commercial publishers. cess’ by permitting other researchers to examine in depth the published work and validate, or contradict, the conclusions. Acknowledgements • Access: In the case of non-OA electronic publishing, a The author has benefited from discussions over many years researcher has to access separately (with identifier and pass- with key people in OA. This short article has benefited from word provided upon payment of the annual subscription) the discussions with Heather Weaver of CCLRC. databases of publications of each publisher to obtain information. In the case of ‘gold’ OA publishing a researcher has Please contact: to access separately the open databases of publications of Keith G. Jeffery, Director IT, CCLRC and ERCIM president each publisher to obtain information. In both of these cases E-mail: [email protected] the user interface is different from publisher to publisher. In

17 EUROPEAN SCENE: OPEN ACCESS

Publish or Perish — Self-Archive to Flourish: The Green Route to Open Access

by Stevan Harnad

Europe is losing almost 50% of the potential return on its research investment until research funders and institutions mandate that all research findings must be made freely accessible to all would-be users, webwide.

It is not the number of articles published proposed to adopt a policy requiring UK tion is worth more than an increase from that reflects the return on Europe’s researchers to deposit, on their univer- 30 to 31; most articles are in the citation research investment: A piece of research, sity’s website, the final author’s draft of range 0-5.) Taking only the most conser- if it is worth funding and doing at all, any journal article resulting from vative low-end of this range ($50), must not only be published, but used, RCUK-funded research. The purpose of updating by about 170% for inflation applied and built-upon by other the proposed policy would be to max- from 1986-2005 and converting to researchers, worldwide. This is called imise the usage and impact of UK Euros, this would yield 73 Euro as the ‘research impact’ and a measure of it is research findings by making them freely marginal value of a citation to its author the number of times an article is cited by accessible on the web (‘open access’) for today. Self-archiving, as noted, increases other articles (‘citation impact’). any potential users in the UK and world- citations by 50%+, but, as also noted, wide who cannot afford paid access to only 15% of the articles being published In order to be used and built upon, an the published journal version. How are being self-archived today. Readers article must first be accessed. A pub- would a similar policy maximise the can calculate for their own respective lished article is accessible only to those return on Europe’s public investment in countries a conservative estimate (50% researchers who happen to be at institu- research? citation increase from self-archiving at tions that can afford to subscribe to the 73 Euro per citation for 85% of their own particular journal in which it was pub- It is not possible to calculate all the ways country’s current annual journal article lished. There are 24,000 research jour- in which research generates revenue. A output) of the total annual loss of rev- nals in all today, across all research good deal of it is a question of probability enue to their country’s researchers for fields, worldwide, but most institutions and depends on time: Although everyone not having done (or delegated) the few can only afford a small fraction of them. thinks of an immediate cure for cancer or extra keystrokes per article it would have In paper days, authors used to supple- a cheap, clean source of energy as the taken to self-archive their final drafts. ment this paid access to their articles by kind of result we hope for, most research mailing free reprints to any would-be progresses gradually and indirectly, and But this impact loss translates into a far users who wrote to request them. The the best estimate of the size and direction bigger one for their country’s tax-paying online age has now made it possible for of its progress is its citation impact, for public, if we reckon it as the loss of authors to provide limitless free ‘eprints’ that reflects the degree of uptake of the potential returns on their annual research by ‘self-archiving’ electronic versions of research results by other researchers, in investment: If a country invests R billion their own final drafts on their own insti- their own subsequent research. Citation Euros in its research, this translates into tutional websites for all potential users impact is accordingly rewarded by uni- the loss of 50% x 85% = 42.5% or close webwide who cannot afford the journal versities (through salary increases and to R/2 billion Euros’ worth of potential version. promotion) as well as by research-fun- citation impact simply for failing to self- ders (through grant funding and archive it all. It is as if someone bought R The online-age practice of self-archiving renewal); it is also rewarded by libraries billion Euros worth of batteries and lost has been shown to increase citation (through journal selection and renewal, 42.5% of their potential usage simply for impact by a dramatic 50-250%, but so far based on the journal’s average citation failing to refrigerate them all before use. only 15% of researchers are actually ‘impact factor’). Counting citations is a And that is without even considering the doing it. Yet two recent UK international natural extension of the cruder measure wider loss in revenue from the loss of surveys have found that 95% of authors of research impact: counting publications potential practical usage and applica- would self-archive – but only if their themselves (‘publish or perish’). tions of each nation’s research findings research funders or their institutions in Europe and worldwide, nor the still require them to do so (just as they already If citations are being counted, it is natu- more general loss to the progress of require them to ‘publish or perish’). The ral to ask how much they are worth. human inquiry. solution is accordingly obvious: For the United States in 1986, Diamond The solution is obvious, and it is the one After lengthy deliberations first initiated estimated the marginal dollar value of the RCUK is proposing: to extend in 2003 by the UK Parliamentary Select one citation as ranging from $50-$1300 research’s existing universal ‘publish or Committee on Science and Technology, (US), depending on field and number of perish’ requirement to ‘publish and also Research Councils UK (RCUK) have citations. (An increase from 0 to 1 cita- self-archive your final draft on your

18 institutional website’. Over 90% of jour- Links: nals are already ‘green’ on author self- • Ulrich’s periodical directory: http://www.ulrichsweb.com/ archiving; two international author sur- • Statistics provided by the Association of Research Libraries: veys report that over 90% of authors will http://fisher.lib.virginia.edu/arl/ comply; and the actual experience of the • Self-Archiving FAQ: http://www.eprints.org/openaccess/self-faq/ five institutions that have so far already • Institutional Archives Registry: adopted such a requirement (CERN, http://archives.eprints.org/eprints.php?action=browse University of Southampton ECS, • Comparing the Impact of Open Access vs. Non-OA Articles in the Same Journals, Queensland University of Technology, D-Lib Magazine, June 2004: http://eprints.ecs.soton.ac.uk/10207/01/06harnad.html University of Minho, University of • Swan, A. (2005) Open access self-archiving: An Introduction. Technical Report, Zurich) tends to bear this out. JISC, HEFCE: http://eprints.ecs.soton.ac.uk/11006/ • Recommendations for UK Open Access Provision Policy: The time for Europe to close its own http://www.ecs.soton.ac.uk/%7Eharnad/Temp/UKSTC.htm 50%-250% research impact gap is • UK Research Councils’ Position on Access to Research Outputs: already well overdue. All of Europe http://www.rcuk.ac.uk/press/20050628openaccess.asp should immediately follow the UK • Effect of open access on citation impact: a bibliography of studies: model, adopting the web-age extension http://opcit.eprints.org/oacitation-biblio.html of “publish or perish” policy to: “publish • Carr, L. and Harnad, S. (2005) Keystroke Economy: A Study of the Time and and self-archive on the web.” This tiny Effort Involved in Self-Archiving: http://eprints.ecs.soton.ac.uk/10688/ and very natural evolutionary step will • Journal Policies - Summary Statistics So Far: http://romeo.eprints.org/stats.php not only be of enormous benefit to Europe’s researchers, its institutions, its Please contact: funders, and its funders’ funders (ie, the Stevan Harnad, American Scientist Open,Access Forum, tax-payers), but it will also be to the col- Department of Electronics and Computer Science, University of Southampton, UK lective advantage of global research E-mail: [email protected] progress and productivity itself, and the http://www.ecs.soton.ac.uk/~harnad/ object of emulation worldwide.

The Golden Route to Open Access by Jan Velterop

Publishing research results is part and parcel of doing research. Without publishing it, one can just as well not do the research. Publishing is not just an option for a serious scientist, but more or less compulsory, albeit to a degree a ‘social compulsion’ - it is "the right thing to do". Often enough, though, it is an inescapable requirement for those who want to make a career in science. Without having a good number of publications to your name, you will find it difficult to make promotion, qualify for tenure, obtain funding for further projects, and get the acknowledgement and recognition most scientists crave. The slogan ‘publish or perish’will sound familiar.

Given this, it is quite remarkable that ever, defrays a very small proportion of gence of the discipline in question. doing research and subsequently pub- the overall cost of publishing. ‘Dissemination’ speaks for itself and is lishing the results have been regarded as the element most directly influenced — mostly separate processes, taking place ‘Publishing’ is quite a loose and ill-delim- improved — by open access. in separate worlds. And it is perhaps ited term which in the context of science ‘Information’ refers to the actual transfer even more remarkable that to an over- and scholarship comprises a number of of data or knowledge contained in a scien- whelming degree the whole process of ‘ations’: registration, certification, dis- tific article; from researcher to researcher, publishing has hitherto been financed by semination, information, preservation, but also from researcher to student and on contributions (vicariously, via libraries) and compensation. ‘Registration’ means occasion directly to the general public. from readers. I say ‘to an overwhelming recording that the research has taken ‘Preservation’ means proper archiving degree’ because it is not quite so that the place, by whom, when, where, and the and ensuring that the material will be process is entirely financed by readers, like, and ensures proper acknowledge- accessible and usable in the future, which as there is, in some disciplines, a small ment and citation. ‘Certification’ means is considered quite a challenge for elec- contribution from authors in the form of that it has passed the filter of peer-review tronic material. And finally ‘compensa- page charges. This contribution, how- and thus conforms to the standards of dili- tion’, which denotes the fact that as a

19 EUROPEAN SCENE: OPEN ACCESS

researcher, having published as expected needs to be reconsidered. After all, when goal of open access is worth overcoming by one’s institution and funding body, one articles are openly and freely available, those challenges, though. In order to help can avoid perishing as a scientist (though the incentive for the reader (vicariously, make the transition, at Springer we have actually thriving requires a bit more, such the library) to pay for subscriptions or decided to leave the choice to authors as citations to one’s articles). licences is materially diminished. Only (and their institutions and funders). They financing the system by contributions can, once their article has been accepted If one looks at these ‘ations’, it is striking from authors (vicariously, from institu- for publication after peer review, opt to that most are of much more importance to tions or funding bodies), who have a have it published in the traditional way, the authors of the material than to prospec- very strong incentive to have their arti- and disseminated via subscriptions, or tive readers. Whether a given article is cles published, makes open access pub- opt to have it published with immediate published or not will hardly ever register lishing economically feasible and robust. and full open access. The scheme called with readers. There are even voices who This has come to be known as the Springer Open Choice, applies to all the say that what readers need most are arti- Golden Route to open access. It makes 1300 or so journals that the company cles that are rarely ever published: nega- sense if one considers that in the end it is publishes, and it is hoped that it provides tive results. For the author, however, pub- neither readers nor authors who pay for an opportunity to make a smooth transi- lishing research results is really part of the system anyway, but academic institution from traditional publishing to Gold completing the research process and of tions and funders, either via subscrip- Open Access publishing, at the pace that utmost importance, hence the adage ‘pub- tions — having no open access — or via the scientific community is comfortable lish or perish’ and not ‘read or rot’. article publishing charges, having open with, and that it will be followed by other access and all its benefits. publishers. A few of those other estab- As said, open access to research articles lished publishers have recently instituted does potentially enhance many of the Unfortunately, making the transition is a similar choice model for a small num- things that are important to authors: dis- fiendishly difficult. Most publishers have ber of their journals, perhaps indicating semination, and with it visibility, the therefore, hitherto, stayed away from the that the idea may be catching on. chance of being cited, information and Gold Route to open access, and a few, the chance of influencing ideas, and even very few, new ones have fully embraced Link: preservation because wide distribution the model and are trying to build their http://www.springer.com/openchoice of the material provides some ‘safety in entire business on it. None of those have numbers’. so far been able to make it work econom- Please contact: ically, perhaps demonstrating the Jan Velterop, However, open access means that the formidable difficulties and challenges a Director of Open Access, Springer traditional way of financing publishing transition to open access presents. The E-mail: [email protected]

ERCIM Statement on Open Access

ERCIM researchers have an interest in Open Access both as producers and consumers of research publications, and as developers of technology to enable and sustain open access.

Recognising the inability of research libraries to meet the • the provision of open access should be made as cost-effec- costs of sustaining their collections, and participating actively tive as possible in the development of appropriate technology, ERCIM has • the provision of open access carries also the responsibility followed with interest the developments in Open Access from for curation of the digital material including cataloguing, the Budapest Declaration through the Bethesda Declaration to archiving, reproducing, safekeeping and media migration. the Berlin Declaration and events since. ERCIM member ERCIM has for many years made available digitally its publi- organisations have been involved in dialogue with national cations and other materials. ERCIM pioneered a pilot project libraries, research funding agencies, commercial publishers, demonstrating homogeneous access to heterogeneous techni- learned societies and government departments. ERCIM sup- cal reports. ERCIM has many years experience in the tech- ports the following principles: nology through the DELOS projects and Network of • research that is funded by the public via government agen- Excellence (http://www.delos.info), and is at the leading edge cies or charities should be available freely, electronically at integrating appropriate open access technology with GRIDs the point of use via the DILIGENT project (http://www.diligentproject.org). • other research should be made equally available subject Individual ERCIM organisations have researched many only to confidentiality required by commercial, military, aspects of the technology required for open access. security or personal medical constraints • quality assurance of research publications must be contin- It is now agreed that the member organisations of ERCIM ued through rigorous peer review which do not already have an open access policy will adopt • associated with research publications, research datasets and these principles and implement them. software should be equally openly available

20 Managing Licenses in an Open Access Community by Renato Iannella

A new project from National ICT Australia (NICTA) and Queensland University of Technology (QUT), called ‘Open Access to Knowledge (OAK), aims to address the emerging needs of the open access community in licensing content.

In particular, OAK will be developing a tics. For example, in the CC Sampling Most license management systems today set of legal requirements and generic licenses the legal code is clear that you are focussed on the distribution of com- licenses that can be used to negotiate and cannot use the work for advertising but mercial and consumer-oriented content, transact (ie share) digital content in an there is no corresponding constraint such as music to mobile phones. The online environment. Technically, the semantics in the CC REL. motive of the OAK project is to investi- OAK project will develop robust Rights gate the legal, semantic, and technical Expression Language (REL) models and One of the major technical hurdles for issues of licensing content in the creative Profiles of the machine-readable Open the CC licenses is the lack of extensibil- industry communities. That is, commu- Digital Rights Language (ODRL). OAK ity of its machine-readable REL. For nities that support a high level of sharing will implement technological mecha- example, a recent report from the UK (eg research, science, and education) nisms to improve open access and man- Common Information Environment without a strict requirement for enforce- agement through the application of these (CIE) on the use of Creative Commons ment of content usage, but still requiring license protocols and services to existing in the public and education sectors raised intellectual property rights to be main- digital repositories. a number of areas where the CC licenses tained, honored, and managed. This pro- lacked support, including: ject will be of immediate benefit to these The Creative Commons (CC) has pro- • geographic restrictions communities in that it will increase the vided worldwide interest in simple • sector restrictions ability to access a vast array of content licensing of content for the open access • third party material (including limited and research material. In an environment communities. The range of CC license duration) where access to knowledge is increas- templates have addressed the legal and • no endorsement clause. ingly important to quality of life and community awareness needs for content career advancement this will provide an licensing. The mapping of the standard In general there are no mechanisms for important resource to the broader com- licenses to other countries legal regimes CC licenses to be tailored to specific munity of knowledge consumers. has ensured the international impact of needs of some communities, which CC. However, there are still some legal could impact on the uptake and broader Links: and technical issues that remain a chal- use of CC style licenses. The ODRL ODRL: http://odrl.net lenge and these will be investigated in Initiative and the Creative Commons CC: http://creativecommons.org the OAK Project. have jointly developed a Profile of the CIE: http://www.common-info.org.uk/ CC REL semantics that have been Some of the legal issues are compounded mapped into the more expressive ODRL Please contact: by the lack of technical solutions. For REL. This Profile does enable greater Renato Iannella, Program Leader example, the need to keep the licenses extensibility and provides semantic National ICT Australia (NICTA) attached or associated with the content at structures to extend the reach of the tem- Tel: +61 7 3000 0484 all times is difficult to implement generi- plate licenses. E: [email protected] cally across all media types. And the fail- http://nicta.com.au ure of this at any point in the distribution The OAK Project will build on the of the content will breach the license emerging new intellectual property conditions, and may result in licensed rights model being developed for ODRL content being re-distributed without Version 2.0 that provides new features proper knowledge of the license condi- (such as Duties and Prohibitions) and tions. supports a wider range of License types. Another objective of this project will be The CC REL is a compact rendering of to investigate software solutions to sup- the semantics of the legal licenses. In port the ODRL Profiles developed for most cases the REL captures the broad open content repositories. These will license conditions, such as the right to include protocols to support negotiation produce derivatives, or prohibits com- of licenses between parties. mercial usage. However, in some cases, the REL does not capture these seman-

21 EUROPEAN SCENE: OPEN ACCESS

W3C at the Forefront of Open Access

by Rigo Wenning Links The Web can be seen as one of the preconditions of today’s discussion about Background Documents on OA Open Access. The World Wide Web Consortium (W3C), including its Members, is •Self-archiving FAQ: very conscious about this role of the Web. To give an example, the W3C, from its http://www.eprints.org/self-faq/ earliest days, has used Web technologies to provide full access to the W3C •Bibliography of OA Advantage: Recommendations and Working Drafts and much other information on the W3C’s http://opcit.eprints.org/oacitation- Web site. biblio.html •American Scientist Open Access All W3C Recommendations and norma- After three years of lengthy and controver- Forum: tive documents are published under the sial discussions, in 2004 the W3C adopted http://amsci-forum.amsci.org/ very liberal W3C document license. This its Patent Policy, a landmark innovation in archives/American-Scientist-Open- license allows people to re-use the docu- the area of standardization. While most Access-Forum.html ment content in all kinds of innovative Specification Developing Organizations ways, on condition that the initial (SDO’s) have adopted a RAND-scheme Recommendation and its attribution to (reasonable and non-discriminatory Policy W3C are not altered. The limitation terms), W3C was the first SDO to adopt a •Budapest: related to document changes is due to the regime of royalty free and non-discrimina- http://www.soros.org/openaccess/ normative character of the W3C tory licensing terms for every patent read.shtml Recommendations, since they represent essential to the implementation of a W3C •Bethesda: a consensus of the community. The pro- Recommendation. This was a major step http://www.earlham.edu/~peters/fos/ hibition to change and create derivatives to help W3C Recommendations to get the bethesda.htm of the W3C Recommendation protects most widespread use and recognition. •Berlin Declaration: this consensus. W3C additionally has a While the innovative Patent Policy created http://www.zim.mpg.de/openaccess- liberal software license that allows the several issues in business procedures for berlin/signatories.html W3C open source code to be altered, W3C as well as for its Members, today we •OECD: contributed to, and incorporated in either are already seeing other SDOs copy the http://www.oecd.org/document/15/0,23 open-source or commercial software. model and numerous discussions sparking 40,en_2649_201185_25998799_1_1_ up elsewhere. 1_1,00.html Patent Policy •UK House of Commons Science and In the early years of W3C’s work on Accountability to the Public Technology Select Committee: Web standards, innovation arose out of a The W3C is also conscious that not every http://www.publications.parliament.uk/ combination of community-wide collab- individual can contribute by becoming a pa/cm200304/cmselect/cmsctech/399 oration on open standards and fierce W3C Member. Therefore the W3C has /3990 competition in implementation of those developed a very open process to accom- standards. Patents were not initially modate views from the public at large. •RCUK: identified as a barrier to innovation or Those contributions and comments have http://www.rcuk.ac.uk/access/ interoperability because no one was to be taken into account by W3C Working statement.pdf aware of any patent claims asserted to Groups. The Working Groups are block standards-based interoperability. expected to respond to comments, and Policies, Registries and Directories However, as the Web became more com- document decisions about whether they •Overview: mercially prominent and the number of accept the suggestion or not. The simplic- http://www.openarchives.org software and business process patents ity of the Working Groups’ email-based •Institutional Archives Registry: increased, some patent holders sought to feedback approach is contributing greatly http://archives.eprints.org/eprints.php require license payments as a condition to the reach of W3C technologies. Those •Institutional Self-Archiving Policy of implementing Web standards. In some comments and the responses are publicly Registry cases, these patent holders had also par- archived for accountability and for infor- http://www.eprints.org/signup/fulllist.php ticipated in the development of those mation. Furthermore, today W3C tech- standards. Based on its experience, the nologies can be used in all kinds of lan- •Journal Self-Archiving Policy W3C community came to the conclusion guages, written vertically or from right to Directory http://romeo.eprints.org/ that it is essential to have a clear patent left, such as in Chinese or Arabic. This •‘Gold’ open-access publication policy governing standards develop- aim to include every community makes channels: http://www.doaj.org/ ment. The policy W3C has adopted was the scale of the Web truly global. •Self-archive repositories: designed to assure the continuation of http://www.eprints.org/self-faq/ the fundamental dynamics of innovation Please contact: http://archives.eprints.org/ and interoperability that made the Web Rigo Wenning, W3C http://openaccess.eprints.org/ successful. E-mail: [email protected]

22 Cream of Science

by Wouter Mettrop

JISC surveys of international author In the Netherlands, a national project has started that aims to make available the Open Access scientific output of all Dutch scientific organizations according to the Open Archives •Swan, A. and Brown, S. (2004) Initiative protocol. ISC/OSI JOURNAL AUTHORS SURVEY Report. Technical Report, On 10th May 2005 the president of the showed that authors often do not know JISC, HEFCE: Netherlands Academy of Arts and that signing the author-publisher agree- http://eprints.ecs.soton.ac.uk/11002/ Sciences (KNAW) Prof. dr. Frits van ment means that they give away their •Swan, A. and Brown, S. (2005), Open Oostrom launched de website copyright to a commercial publisher. access self-archiving: An author www.creamofscience.org. It was the DARE partners dealt differently with study. Technical Report, External result of months of very hard work for publication dated from 1998. Some only Collaborators, JISC, HEFCE: the project ‘Keur der Wetenschap’ recorded the metadata from publications http://eprints.ecs.soton.ac.uk/10999/ (Cream of Science) at all Dutch univer- of that period whereas others recorded all •Swan, A., Needham, P., Probets, S., publications, often at request of Muir, A., Oppenheim, C., O’Brien, A., the scientists. Hardy, R. and Rowland, F. (2005), Delivery, Management and Access On behalf of NWO (Netherlands Model for E-prints and Open Access, Organisation for Scientific Journals within Further and Higher Research) the CWI Library takes Education. Technical Report, JISC, part in DARE. It will also host HEFCE: the output of the other NWO http://eprints.ecs.soton.ac.uk/11001/ institutes, although each institute will be responsible for their own input. The CWI-IR can be found Copyright status of publishers through the Keur website or at •http://romeo.eprints.org/stats.php www.darenet.nl, where also some CWI reports of the past OA citation impact years can be found. In the near •http://opcit.eprints.org/oacitation- future an interface will be biblio.html Prof. Dr. Frits van Oostrom launches installed on CWI-IR, which can be •http://citebase.eprints.org/isi_study/ Cream of Science (Photo courtesy of Theo searched separately. Gradually the publi- Koeten Photography). cations of all past and present CWI Effect of OA on Journal researchers will be put in the repository. Subscriptions •Swan, A. and Brown, S (2005) Open sity libraries, KNAW and also at CWI. CWI emphasizes that participation in access self-archiving: an author They were each asked to put the com- Keur der Wetenschap is very useful as a study. plete scientific output of about ten active start for an Institutional Repository. http://www.jisc.ac.uk/uploaded_docu scientists in a repository. This project is Eventually the importance of an IR ments/Open%20Access%20Self%20 part of the national project DARE according to the OAI protocol is to be Archiving-an%20author%20study.pdf (Digital Academic Repositories). DARE found in the possibility of subject-based aims at stimulating so called Institutional ‘harvesting’ by international service Repositories (IR) in the Netherlands. providers. Repositories / Systems These IRs aim to make available the sci- •arXiv: http://arxiv.org/ entific output of all Dutch scientific orga- Link: •ePubs: http://epubs.cclrc.ac.uk/ nizations both digitally and according to http://www.creamofscience.org •GNU eprints software: the international OAI - Open Archives http://www.eprints.org/ Initiative - protocol. Please contact: •DSpace: http://www.dspace.org/ Wouter Mettrop, CWI Copyright was an important issue for Tel: +31 (20) 5924042 •Fedora: http://www.fedora.info/ many participating libraries and authors. E-mail: [email protected] •Citebase: http://citebase.eprints.org/ Is it proper to publicly present publications without the author’s permission? DARE investigated this matter and it turned out that all material dated before 1998 can be admitted to repositories without legal restrictions. Research also

23 SPECIAL THEME: Emergent Computing

Emergent Computing Introduction to the Special Theme

by Heather J. Ruskin and Ray Walshe

Emergent Computing has become more topical and workshops aimed at Emergent Computing, in the last few years and has recently been cate- Emergent Properties, Complexity and Co- gorised as a research field in its own right. The Evolution, to name but a few. The European history of the field lies primarily in Artificial Commission has also recently published the Intelligence, Numerical Methods and fifth call for proposals under the ‘information Complexity Theory all of which have con- society technologies’ (IST) priority of the Sixth tributed in no small part to Emergent Framework Programme (FP6), which falls Computing. Emergent Computing is some- under the seminal programme of ‘integrating times described as ‘highly complex processes and strengthening the European Research arising from the cooperation of many simple Area’. Significantly, a specific targeted processes’, ie high-level behaviour resulting research initiative, included in FP6, is the from low- level interaction of simpler building research heading ‘Simulating Emergent blocks. One example of emergent behaviour Properties in Complex Systems’. that has been studied over recent years is that of ‘flocking’. In a ‘flock of birds’, the flock is a Emergent Computing is influenced by and bor- dynamic entity consisting of hundreds (or thou- rows heavily from other disciplines and one of sands) of individuals. The flock constantly the most prolific symbioses has been that with changes shape and direction but maintains Biology. Systems Biology has provided many overall cohesion. Using cellular automata and bio-inspired approaches, (Biomimetic models treating each bird in the flock as an autonomous and methods), giving rise to emergent proper- agent, with simple local rules controlling agent ties. Neural networks, (from the biological behaviour relative to closest neighbouring ‘neuron’ operation), which form the basis of birds, a system can be constructed where the many clustering, classification and learning overall behaviour of the collective agents systems, provide one of the earliest examples. reflects the behaviour of the real flock. Emergence of conditions, favourable to viral invasion and disease spread, has been modelled The Emergent Computing paradigm both using cellular automata and autonomous agent explores and relies upon biologically and methodologies. The behaviour patterns of socially inspired systems, in which complex swarming bees, flocking birds and schools of behaviour at the global level emerges in a non- fish, have generated so-called swarm technolo- linear manner from large numbers of low-level gies, used in complex dynamic systems prob- component interactions. Building software sys- lems such as scheduling, optimisation and tems, using this component methodology, space exploration. Similarly, a portfolio of offers many advantages for solving complex mathematical and numerical methods, statisti- problems, since the algorithmic complexity is cal and probabilistic reasoning, have found achieved through software that is simple and applications in learning systems for adaptation flexible compared to conventional software and prediction, linguistics, intelligence and development techniques control.

Building systems, with behaviour more than As the field of emergent computing progresses the sum of its parts, attracts methodologies and and computational power allows us to look techniques from a number of disciplines, so we more closely at the individual components, to include as a broader definition of the topic, the manipulate them more efficiently, and to com- mathematical and computational techniques bine and compare readily existing techniques that underpin the area of Emergent Computing. with new and hybrid methodologies, evolution- These include examples such as classifier sys- ary complexity seems almost within our grasp. tems, neural networks, biological immune sys- As size and sophistication increase of what can tems, autocatalytic networks, adaptive game be modelled, useful insight may be gained on theory, chaos theory, general nonlinear systems systems and applications, with scales ranging and artificial life. from the sub-cellular to the social and beyond.

Recent increased interest in these topics is illus- This issue aims to highlight some of the areas trated by the many international conferences and techniques of current interest internatio-

24 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

nally in Emergent Computing research, ARTICLES IN THIS SECTION through both special theme articles sub- 24 Introduction to the Special Theme 42 Particle Swarm Optimization for the mitted and invited articles from key by Heather J. Ruskin and Ray Walshe, Reconstruction of Permittivity Range researchers. Dublin City University/IUA, Ireland Profiles from Microwave Measurements by Simone Genovesi and Emanuele Invited article: Salerno, ISTI-CNR, Italy The invited articles introduce topics such 26 Swarm-Based Space Exploration by Michael G. Hinchey, Roy Sterritt, 44 Building Blocks from Biology for the as: Christopher A. Rouff, James L. Rash and Design of Algorithms for the • unmanned space exploration using Walt Truszkowski Management of Modern Dynamic intelligent swarm technologies Invited article: Networks • co-evolution of emergent properties in 27 Collective Intelligence and Evolution by Gianni A. Di Caro, Frederick Ducatelle, by Akira Namatame, Akira Namatame, Luca Maria Gambardella, Andrea Rizzoli collective systems (collective intelli- National Defense Academy, Japan Istituto Dalle Molle di Studi sull'Intelligenza gence) Invited article: Artificiale (IDSIA), Manno-Lugano, • evolution of computer programs by 28 Evolving Game-Playing Strategies Switzerland with Genetic Programming artificial selection, using genetic pro- 45 Chemical Programming of Self- by Moshe Sipper, Ben-Gurion University, Organizing Systems gramming, to obtain human-competi- Israel tive game-playing strategies by Jean-Pierre Banâtre, Pascal Fradet and 30 Collaborative Online Development Yann Radenac, INRIA/IRISA, France of Modular Intelligent Agents 46 A Development Model The 21 special theme articles in this by Ciarán O’Leary, Dublin Institute of for Robust Fault-Tolerant Design issue highlight current research trends, Technology, Mark Humphrys and Ray by Andy Tyrrell and Hong Sun, University Walshe, Dublin City University/IUA, Ireland not only within the ERCIM community, of York, UK but more widely across Europe. Included 31 CE-Ants: Ant-Like Agents for Path 48 Emergent Properties of the Human Management in the Next-Generation are topics, such as: Immune Response to HIV Infection: Internet • evolving higher order intelligence and Results from Multi-Agent Computer by Otto Wittner and Bjarne E. Helvik, Simulations evolving classifier systems NTNU, Norway • particle swarm optimisation in by Ashley Callaghan, Dublin City 33 ISEE – A Framework for the Evolution University / IUA, Ireland microwave backscattering data and and Analysis of Recurrent Neural 49 Network Emergence in Immune System metaheuristics in which a decentral- Networks for Embodied Agents Shape by Martin Hülse, Steffen Wischmann, and ized design leads to emergent phenom- by Heather Ruskin, and John Burns, Keyan Zahedi, Fraunhofer Institute for ena Dublin City University, Ireland Autonomous Intelligent Systems – AIS, • fuzzy coalitions in game playing Fraunhofer ICT Group, Germany 50 Agent-Based Modelling of Viral • self-organised and self-optimisation in Infection 34 Emergent Intelligence in Competitive by Dimitri Perrin, Dublin City University / dynamic networks and chemical pro- Multi-Agent Systems IUA, Ireland gramming by Sander M. Bohte, Han La Poutré, CWI, • predictive modelling in financial net- The Netherlands 51 The ‘Decent’ Project: Decentralized Metaheuristics works and statistical modelling frame- 35 Building a Computational Digital by Enrique Alba, University of Málaga / Economy through Interdisciplinary works SpaRCIM, Spain, Research • evolvable hardware using a multi-cel- and Martin Middendorf, University Leipzig, by Petros Kavassalis, University of Crete, Germany lular organism approach Greece, and Konstantin Popov, Konstantin • emergent development in the walking Popov SICS, Sweden 53 Evolutionary Methods make New behaviour of a robot Effective Laser Shapes 37 Models of Multilateral Cooperative by Thomas Bäck and Joost N. Kok, Leiden • agent based modelling of viral infec- Behaviour University, by Milan Mareš Institute of Information tion and network emergence in the The Netherlands immune system Theory and Automation, Academy of Sciences / CRCIM, Czech Republic 54 Neural Net Modelling in Financial • collaborative online development of Engineering of Options 38 Emergent Walking Behaviour in an modular intelligent agents and emer- by Jerome Healy, Longchuan Xu, Maurice Aibo Robot Dixon, Fang Fang Cai, Brian Read and gent intelligence in competitive multi- by Cecilio Angulo, Ricardo A. Téllez and Brian Eales agent systems. Diego E. Pardo, Technical University of Catalonia (UPC) / SpaRCIM, Spain 55 Self-Optimization in a Next-Generation Urban Traffic Control Environment Link: 39 Concept-Based Text Representations by Raymond Cunningham, Jim Dowling, http://bioinformatics.computing.dcu.ie for Categorization Problems Anthony Harrington, by Magnus Sahlgren, SICS, Sweden Vinny Reynolds, René Meier and Vinny Please contact: 40 A Framework for Efficient Statistical Cahill, Trinity College Dublin / IUA, Ireland Modelling Heather J. Ruskin 56 Self-Organized Routing in Mobile by Daniel Gillblad and Anders Holst, SICS, Dublin City University/IUA, Ireland Ad Hoc Networks using SAMPLE Sweden E-mail: [email protected] by Jim Dowling and Stefan Weber, Trinity 41 Rules, Inferences and Robust College Dublin / IUA, Ireland Approximation at Work Ray Walshe by Antonín Dvoﬁák, Vilém Novák, Viktor Dublin City University/IUA, Ireland Pavliska, University of Ostrava, Czech E-mail: [email protected] Republic

ERCIM News No. 64, January 2006 25 SPECIAL THEME: Emergent Computing

Swarm-Based Space Exploration

by Michael G. Hinchey, Roy Sterritt, Christopher A. Rouff, James L. Rash and Walt Truszkowski

Future space exploration missions will exploit intelligent swarm technologies, enabling spacecraft to go where manned missions and traditional spacecraft simply cannot.

Planned NASA missions will exploit and will obtain specific types of data. A Complex Problem new paradigms for space exploration, Some will be coordinators (called lead- New approaches to exploration missions heavily focused on the (still) emerging ers) that have rules that decide the types such as ANTS augur great potential, but technologies of autonomous and auto- of asteroids and data the mission is inter- simultaneously pose many challenges. nomic systems. Traditional missions, ested in and that will coordinate the The missions will be unmanned and nec- reliant on one large spacecraft, are being efforts of the workers. The third type of essarily highly autonomous. They will complemented by missions that involve spacecraft are messengers that will coor- also exhibit the properties of autonomic several smaller spacecraft, operating in dinate communication between the systems of being self-protecting, self- collaboration, analogous to swarms in rulers and workers, and communications healing, self-configuring, and self-opti- nature. with the Earth ground station. mizing in order to assist in the surviv- ability of the mission. Many of these This offers several advantages: the abil- The swarm will form sub-swarms under missions will be sent to parts of the solar ity to send spacecraft to explore regions the control of a ruler, which contains system where manned missions are sim- of space where traditional craft simply models of the types of science that it ply not possible, and to where the round- would be impractical, greater redun- dancy (and, consequently, greater protection of assets), and reduced costs and The operational ANTS risk, to name but a few. mission concept.

ANTS: A Concept Mission ANTS (Autonomous Nano Technology Swarm) is a concept NASA mission. In one of its sub-missions, Prospecting Asteroid Mission (PAM), a transport ship, launched from Earth, will travel to a point in space where gravitational forces on small objects (such as pico- class spacecraft) are all but negligible. From this point, termed a Lagrangian, 1000 spacecraft, which will have been assembled en route from Earth, will be launched into the asteroid belt. It is expected that as much as 60 to 70 percent wants to perform. The ruler will coordi- trip delay for communications to space- of them will be lost during the mission, nate workers, each of which uses its indi- craft exceeds 40 minutes, meaning that primarily because of collisions with each vidual instrument to collect data on spe- the decisions on responses to problems other or with an asteroid during explo- cific asteroids and feed this information and undesirable situations must be made ration operations, since, having only back to the ruler, who will determine in situ rather than from ground control solar sails to provide thrust, their ability which asteroids are worth examining on Earth. to maneuver will be severely limited. further. If the data matches the profile of Because of their small size, each space- a type of asteroid that is of interest, an NASA Goddard Space Flight Center has craft will carry just one specialized imaging spacecraft will be sent to the been collaborating with University of instrument for collecting a specific type asteroid to ascertain the exact location Ulster and Science Applications of data from asteroids in the belt. and to create a rough model to be used by International Corp. (SAIC) on develop- other spacecraft for maneuvering around ing techniques and new self-paradigms, Approximately 80 percent of the space- the asteroid. Other teams of spacecraft which may be applicable to future craft will be workers that will carry the will then coordinate to finish mapping swarm-based missions. In addition, the specialized instruments (eg, a magne- the asteroid to form a complete model. degree of autonomy that such missions tometer or an x-ray, gamma-ray, visi- will possess would require a prohibitive ble/IR, or neutral mass spectrometer) amount of testing in order to accomplish

26 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

system verification. Furthermore, learn- dict the emergent behavior of 1000 The project is currently working on inte- ing and adaptation with the goal of con- agents operating as a swarm, as well as grating existing formal techniques and tinual improvements in performance will the behavior of the individual agents. on building tools to support the inte- mean that emergent behavior patterns Crucial to the success of the mission will grated method. simply cannot be fully predicted through be autonomic properties and the ability the use of traditional system develop- to modify operations autonomously to Links: ment methods. reflect the changing nature of the mis- ANTS website: http://ants.gsfc.nasa.gov sion. For this, the formal specification NASA Software Engineering Laboratory: The Formal Approaches to Swarm will need to be able to track the goals of http://sel.gsfc.nasa.gov Technology (FAST) project aims at the mission as they change and to modify UU Autonomic Systems: devising a formal approach to the devel- the model of the universe as new data http://www.infj.ulst.ac.uk/~autonomic opment and verification of complex comes in. The formal specification will swarm-based systems, using ANTS as a also need to allow for specification of the Please contact: baseline for comparing approaches. decision-making process to aid in the Dr. Mike Hinchey, Director, NASA Software decision of which instruments will be Engineering Laboratory, USA An effective formal method for use in needed, at what location, with what Tel: +1 301 286 9057 the ANTS mission must be able to pre- goals, etc. E-mail: [email protected]

Collective Intelligence and Evolution by Akira Namatame

The mission of collective evolution is to harness the systems of selfish agents to secure a sustainable relationship, so that desirable properties can emerge as ‘collective intelligence’.

Why do colonies of ants work collec- how our social and economic activities evolution in nature. Evolution through tively, and how do they do it so effec- should be organized. natural selection is often understood to tively? One key to answering this ques- imply improvement and progress. If tion is to look at interactions among ants. On the other hand, the fact that selfish multiple populations of species are For the last decade, attempts have been behaviour may not achieve full effi- adapting each other, the result is a co- made to develop some general under- ciency is also well known in the litera- evolutionary process. However, the standing, which has produced the theory ture. It is important to investigate the loss problem to contend with in Darwinian of collective systems, that is, systems of collective welfare due to selfish and co-evolution is the possibility of an esca- consisting of a large collection of agents. uncoordinated behavior. Recent research lating arms race with no end. Competing It is common to refer to the desirable efforts have focused on quantifying this species may continually adapt to each emergent properties of collective sys- loss for specific environments, and the other in more and more specialized tems as ‘collective intelligence’. resulting degree of efficiency loss is ways, never stabilizing at a desirable Interactions are able to produce collec- known as ‘the price of anarchy’. outcome. tive intelligence at the macroscopic level Investigations into the price anarchy that is simply not present when the com- have provided some measures for The Rock-Scissors-Paper (RSP) game is ponents are considered individually. designing collective systems with a typical form of representing the trian- robustness against selfish behaviour. gular relationship. This simple game has The concept of collective intelligence Collective systems are based on an anal- been used to explain the importance of observed in social insects can be ogous assumption that individuals are biodiversity. We generalize a basic rock- extended to humans. In his book, The selfish optimizers, and we need method- scissors-paper relationship to a non- Wisdom Of Crowds, Surowiecki ologies so that the selfish behaviour of zero-sum game with the payoff matrix explores a simple idea that has profound individuals need not degrade the system shown in Table 1. In this triangular situa- implications: a large collection of peo- performance. Of particular interest is the tion, diversity resulting from proper dis- ple are smarter than an elite few at solv- issue of how social interactions should persal by achieving Nash equilibrium is ing problems, fostering innovation, be restructured so that agents are free to not efficient, and the agents may benefit coming to wise decisions, and predict- choose their own actions, while avoiding from achieving a better relationship. ing the future. His counterintuitive outcomes that none would choose. notion, rather than crowd psychology as In particular, we have examined the sys- traditionally understood, provides us Darwinian dynamics based on mutation tem of interactive evolving agents in the with new insights for understanding and selection form the core of models for context of repeated RSP games, by con-

ERCIM News No. 64, January 2006 27 SPECIAL THEME: Emergent Computing

optimal outcomes: (0,1) (1,2) (2,0) (1,0) (2,1) (0,2). Therefore each agent learns to behave as follows: win three times and then lose three times. In this way, the agents succeed in collectively evolving a robust learning procedure that leads to near-optimal behaviour based on the principle of give and take.

The framework of collective evolution is distinguished from co-evolution in three Table 1: The generalized rock-scissors- aspects. First, there is the coupling rule: a Figure 1: The state diagram of the paper game ( λ ≥ 2 ). deterministic process that links past out- strategy choices between two agents. comes with future behaviour. The second aspect, which is distinguished from sidering a population of agents located individual learning, is that agents may or learning by the agents. This approach on a lattice network of 20x20. They wish to optimize the outcome of the joint to collective evolution is very much at the repeatedly play the generalized RSP actions. The third aspect is to describe forefront of the design of desired collec- game with their nearest eight neighbours how a coupling rule should be improved, tives in terms of efficiency, equity, and based on the coupling rules, which are using the criterion of performance to sustainability. Further work will need to updated by the crossover operator. 400 evaluate the rule. examine how collective evolution across different rules, one for each agent, are the complex socio-economical networks aggregated at the beginning into a few In biology, the gene is the unit of selec- leads to emergent effects at higher levels. rules with many commonalities. The tion. However, the collective evolution- game between two agents with the ary process is expected to compel agents Please contact: learned coupling rule becomes a kind of towards ever more refined adaptation, Akira Namatame, stochastic process. The transitions of the resulting in sophisticated behavioural National Defense Academy, Japan outcome are represented as the phase rules. Cultural interpretations of collec- Tel: +81 468 3810 diagram in Figure 1, and they converge tive evolution assume that successful E-mail: [email protected] into the limit cycle, visiting the Pareto- behavioural rules are spread by imitation http://www.nda.ac.jp/~nama

Evolving Game-Playing Strategies with Genetic Programming

by Moshe Sipper

We have recently used genetic programming, wherein computer programs evolve by artificial selection, to obtain human-competitive game-playing strategies for three games: chess, backgammon, and Robocode.

The idea of applying the biological prin- whose basic building blocks are The full list (for backgammon) comprises ciple of natural evolution to artificial designed for the problem at hand. For over 20 such programmatic elements. systems, introduced over four decades example, when we evolved backgam- ago, has seen impressive growth in the mon-playing programs the list of ele- The main mechanism behind genetic past few years. Evolutionary algorithms ments from which programs could be programming is that of a generic evolu- have been successfully applied to constructed included: tionary algorithm (which, in turn, is numerous problems from different inspired by nature), namely, the repeated domains, including optimization, auto- • Player-Exposed(n): Test whether the cycling through four operations applied matic programming, circuit design, player has exactly one checker at to the entire population: evaluate, select, machine learning, economics, immune board location n crossover, and mutate. Start with an ini- systems, ecology, and population genet- • Player-Blocked(n): Test whether the tial population of randomly generated ics, to mention but a few. Our group player has two or more checkers at programs composed of the program ele- focuses on the evolutionary methodol- location n ments for the problem at hand (eg, ogy known as genetic programming. • Enemy-Exposed(n): Test whether the backgammon); this random population is enemy has exactly one checker at known as generation zero. Each individ- In genetic programming we evolve a board location n ual is then evaluated in the domain envi- population of computer programs, • Sub(F, F) : Subtract two real numbers. ronment and assigned a fitness value rep-

28 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing resenting how well the individual solves Figure 2: the problem at hand. In the backgammon Generic programming example fitness can by assigned by hav- flowchart. M is the population ing each individual play against some size and Gen is the generation external (perhaps commercial) program counter. The termination that acts as a teacher, or by having the criterion can be the individuals play against each other, the completion of a fixed number latter referred to as coevolution. of generations or the discovery of a good-enough Being randomly generated, the first-gen- individual. eration individuals usually exhibit poor performance. However, some individuals are better than others, ie, (as in nature) variability exists, and through the mechanism of natural (or, in our case, artificial) selection, these have a higher probability of being selected to parent the next generation. Once individuals have been proba- bilistically selected in accordance with fitness, two genetically inspired operators are applied: crossover, which takes two individual programs, ‘cuts’ each in two, and then combines the four resulting pieces to create two new viable offspring; and mutation, which takes one program and mutates it, ie, changes it in some random manner that results in a different previous machine players—quite a win now and again, ie, the evolved (legal) program. Thus we have gone significant improvement. program definitely holds its own. through one evaluate-select-crossover- • Chess (endgames): We evolved play- • Robocode: A simulation-based game mutate cycle, known as a generation. ers able to play several types of in which robotic tanks fight to destruc- Repeating this procedure, cycling through endgames. While endgames typically tion in a closed arena. The program- several generations, may ultimately lead contain but a few pieces, the problem mers implement their robots in the to good solutions to the given problem, in of evaluation is still hard, as the pieces Java programming language and may our case an artificial player able to hold its are usually free to move all over the submit them to a central web site own against human or machine players. board, resulting in complex game where online tournaments regularly trees—both deep and with high take place. GP-Robocode’s best score Other than backgammon we tested our branching factors. We pitted our was first place of 28 in the interna- evolutionary approach on two other evolved GP-EndChess players against tional league, with all other 27 players games — chess, and Robocode — two very strong programs: MASTER having been written by humans. obtaining excellent results: (which we wrote based on consultation • Backgammon: We evolved full- with high-ranking chess players) and These results recently earned us a Bronze fledged players for the non-doubling- CRAFTY—a world-class chess pro- Medal in the 2005 Human-Competitive cube version of the game. Pitted gram, which finished second in the Awards in Genetic and Evolutionary against Pubeval, a standard benchmark 2004 World Computer Speed Chess Computation competition, where the program, our top programs won 62% Championship. GP-EndChess was objective was to have evolution produce of the matches, 10% higher than top able to draw most of the time and even results competitive with humans.

Links: Author’s website: http://www.moshesipper.com/ Human-competitive awards: http://www.human-competitive.org/ GP-Robocode wins first place: http://robocode.yajags.com/20050625/ haiku-1v1.html

Please contact: Moshe Sipper, Ben-Gurion University, Israel Figure 1: E-mail: [email protected] Robocode interface.

ERCIM News No. 64, January 2006 29 SPECIAL THEME: Emergent Computing

Collaborative Online Development of Modular Intelligent Agents

by Ciarán O’Leary, Mark Humphrys and Ray Walshe

There is a growing consensus that natural intelligence is likely to be composed of multiple diverse algorithms and knowledge representation formats, with hundreds (or thousands) of specialised subsystems collaborating in some network (or ‘Society of Mind’). This is in contrast to the ‘one size fits all’ approach of many popular algorithms in Artificial Intelligence. If such diverse, extremely hybrid Figure 1: The basic World-Wide-Mind artificial agents are to be built, many authors will need to be involved, each with architecture. Client queries both world a different area of specialization. And if the future of AI is one of multiple authors and mind services for state and action and multiple laboratories building vast collaborative systems, how is this work data, passing the required data between to be organized? the services.

The World-Wide-Mind project being the best known). These are natu- (w2mind.org) proposes that AI research rally inclined towards dividing the agent should be organized online, using an mind into components and integrating the agreed, stripped-down, simplified proto- components to create whole minds. Such col for communication between different approaches, generally referred to as subcomponents of the artificial minds. ‘Society of Mind’ approaches after This is related to large online ‘open Minsky’s famous book, recognize the source’ collaborative projects, but differs importance of masking the implementa- from them in that it is not proposed that tion details of individual components the components actually have to be open from other components, while retaining a source. In many cases, users of a third public interface through which they can party’s subsystem (or ‘sub-mind’) need interact. This reflects the modern Figure 2: A Society of Mind on the World- not understand how it works, but only approach to software engineering, where Wide-Mind. The MindM service developer how to use it as a component in the larger modular development is seen to support creates a switch that selects the ‘mind’ they are constructing. By keeping reuse, maintainability, extensibility and appropriate mind service for a given the framework as simple as possible and in general the creation of software sys- state, delegating responsibility to that by populating the Web with multiple tems where no individual is required to service. algorithms, the task of the agent devel- understand the entire system. oper is reduced to selecting the algorithms and plugging them into the agent A natural extension of the modular mind. What results are new and novel approach in software engineering is the agent minds composed of many diverse distributed modular approach. The Society of Mind Markup Language components, designed and implemented Middleware standards such as CORBA (SOML) is the protocol of the World- by multiple, loosely coupled collabora- and Web Services have provided devel- Wide-Mind (WWM), in much the same tors. opers with the means to connect modules way as HTTP is the protocol of the over the Internet without knowing any- World-Wide-Web. The WWM defines The project was started in 2001 by Dr thing about remote modules apart from three types of entities, each of which Mark Humphrys at Dublin City their URL and interface. Although mid- interacts with the others using this proto- University, as an attempt to facilitate the dleware platforms such as JADE (Java col. integration of the many diverse compo- Agent Development Environment) pro- nents of agent minds into whole minds. vide support for multiple interacting A world service on the WWM represents Though much fascinating research is per- autonomous agents, there has been little some sort of problem, eg a robot in a formed in focused subdomains of AI, or no attention directed towards how the complex environment, or a creature in such as perception, vision, navigation distributed modular approach to soft- some virtual reality. The world service and language processing, there has tradi- ware engineering can support the modu- presents a public interface that can be tionally been less effort directed towards lar implementation of whole agent queried for the state of the world (such as the problem of integrating all these com- minds, or how the integration of inde- the sensory inputs of the creature) and ponents into single, whole agents. Within pendent remote components can further can also be sent an action that the crea- the past two decades the animat and our understanding of the diversity ture should execute. behaviour-based communities have con- required for artificially intelligent tributed several modular agent architec- agents. A mind service represents some action- tures (Brooks’ subsumption architecture producing algorithm specifically

30 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

designed for a given world service. The high-level switch that acts as a mind ser- the protocols are in the public domain, mind service will return an action when vice, but simply decides at each timestep anyone can contribute to the project by it is presented with the state of the world. to which of the multiple mind services it developing their own world service, or The implementation details of the algo- will delegate responsibility. Such implementing a mind or MindM service rithm are hidden behind the interface, so switches, named MindM services, facili- for existing worlds. any state maintained by the algorithm is tate the creation of Societies of Mind, transparent to the calling entity. where the many mind services are A key benefit of the framework is that it authored by a diverse set of independent is open to all. Complex minds can be A client is used to plug a mind service researchers. built by integrating the product of into a world service. When launched, the advanced research by professionals with client requests the URLs of the world The technology required to implement the novel algorithms of interested ama- service and mind service from the user. It mind services, world services and clients teurs. The result will be complex then executes a run by querying the is kept deliberately simple, well below arrangements of never-before-seen algo- world for its state and the mind for its the threshold of knowledge required for rithms, possibly bringing us closer to action based on this state, before query- complex distributed systems develop- understanding how our own minds and ing the world again to execute the action. ment. SOML entities interact over the minds of other animals are made so The run iterates through these steps until HTTP, meaning that the creation of a much richer in the presence of widely its natural conclusion (death of the vir- service requires only basic knowledge of diverse components. tual creature, expiration of time permit- Web programming (CGI, Java Servlets, ted for robot), at which point a score is ASP etc). Link: recorded on a publicly accessible score- http://w2mind.org board at the world service. The framework has been tested with several classes of undergraduate students at Please contact: With multiple online mind services, a Dublin City University and Dublin Ciarán O’Leary third-party developer could select a set Institute of Technology. Currently, two Dublin Institute of Technology, Ireland of useful mind services and provide a world services are online, although since Tel: +353 1 402 4718 E-mail: [email protected] CE-Ants: Ant-Like Agents for Path Management in the Next-Generation Internet by Otto Wittner and Bjarne E. Helvik

In general, today’s Internet can provide only ‘best effort’ connections, with no quality of service (QoS) guarantees. Both new and popular existing applications and services would benefit from connections with QoS guarantees, eg minimum capacity, maximum delay, continuity of service and availability. How can connections in the Next-Generation Internet (NGI) be managed such that desired QoS guarantees can be provided while the robustness of today’s network is maintained? We pursue an approach inspired by the robustness and emergent behaviour seen in swarms combined with stochastic optimization techniques.

One of the fundamental design objectives Resource Control ized planning of resource utilization and of today’s Internet was connectivity and Path Management path management, which significantly robustness. First of all, two terminals con- It is common for network operators to weakens the inherent robustness and nected to the network should be able to use over-provisioning combined with autonomy of the traditional Internet. communicate. Secondly, the network virtual connections realized by multipro- should stay operative and provide reliable tocol label switching (MPLS), as a way Furthermore, when multiple paths exist communication even when a significant of providing sufficiently low packet between two communicating terminals, number of network nodes and transmission losses and delays, load distribution and today’s routing systems by default for- links fail. Hence packet forwarding is con- capacity control. Continuity of service, ward all packets along the path regarded trolled in a fully distributed manner and short downtimes and high availability as having the lowest cost. The exact defi- routing units are more or less independent are obtained by supplementing the tradi- nition of the lowest cost is generally of each other. In this way, the Internet pro- tional Internet restoration techniques decided by the network operator, and is vides connectivity to everybody with a with protection switching based on not influenced by applications and users. high probability. Unfortunately, no QoS MPLS and/or in the underlying optical Hence path management does not handle guarantees or differentiations can be given. layer. These techniques require central- service differentiation. As for MPLS

ERCIM News No. 64, January 2006 31 SPECIAL THEME: Emergent Computing

configurations, cost metrics are set man- pairs while seeking to fulfil a set of close-to-optimal paths. As long as fac- ually and infrequently during centralized requirements (the path objective) pro- tory nodes produce new agents, the mes- planning. Hence it is only possible to vided by the owner/producer of the sage distribution will be maintained. consider long-term variation (ie over agents. In their simplest form, agents may Hence, if traffic conditions change or months to years) in general traffic pat- only be small signalling packets trans- link/node failures occur, the optimal path terns and topology. porting state information. Associating a (and the close-to-optimal path) will be behaviour with them, however, makes updated accordingly. The time scales of Would it be possible to design a robust the system description more intuitive. such updates are governed by the pro- path management system that is compli- Each node manages a database of mes- duction frequency in the factory nodes. ant with the ‘traditional Internet’, that sages that are read and updated by the Finally, any node may initiate a connec- agents, that is, tion set-up by applying the information agents commu- provided by the message distribution, ie nicate asyn- it may regard the message distribution as chronously. a distributed routing table and utilize any Selected nodes of the paths found. (source nodes) provide agent Three path management systems based factory function- on CE-ants are currently being devel- ality, while other oped at Q2S and ITEM. Adaptive nodes (destina- restoration ants simply maintain routing tion nodes) pro- information continuously in a network. vide search Traffic is assumed to be routed (stochas- focus control. tically or along the ‘best’ path) hop-by- All nodes pro- hop by applying the information. On CE-ant-based path management. vide path utiliza- failures, traffic is forwarded along ‘section functional- ond-best’ hops. Primary–backup ants ity. Note that no find two (link and/or node) disjoint paths operates on short and medium timescales single node or component is significantly between source-destination pairs, mean- (minutes to hours), provides controlled more important than any other. There is ing they provide 1:1 protection. Back-up QoS differentiation of services or service no centralized control and no centralized paths share capacity. P-cycle ants find classes, and enforces operator policies in storage of information. cyclic paths that may be applied as pro- an optimal way? We believe self-orga- tection rings. On link and node failures, nizing systems inspired by the robust- During a single CE-ant agent life cycle, a selected traffic may be switched onto ness and emergent behaviour seen in path is found and reported in the follow- and rerouted around the ring. swarms may have the necessary proper- ing manner: ties to enable such next-generation path 1.After visiting a node (initially the fac- A Java-based prototype of a CE-ant sys- management. tory node) an agent decides which tem has been developed at Q2S and node to visit next by applying a proba- ITEM. A testbed in the EU-project CE-Ants bility distribution. The distribution is BISON will soon be ready, in which CE- At the Centre for Quantifiable Quality of based on the agent’s memory of previ- ants perform monitoring tasks in a sys- Service in Communication Systems ously visited nodes, the path objective, tem based on Click software routers. (Q2S) and the Department of Telematics and the messages (written or updated These prototypes have shown that only a (ITEM) at NTNU, a set of path manage- by other cooperating/competing limited effort is required to realize CE- ment systems is being developed based agents) available in the current node. ant based path management. on a bio-inspired stochastic optimization 2.When reaching its given destination technique known as Cross Entropy ants node, ie completing a path, an agent Links: (CE-ants). The CE-ant system is a dis- evaluates the cost of the path found http://www.q2s.ntnu.no tributed version of Reuven Rubinstein’s and adjusts search focus parameters http://www.item.ntnu.no popular cross entropy method for opti- shared with cooperating agents. mization. CE-ants have been shown to 3.Finally, the agent backtracks along the Please contact: find near-optimal solutions to NP-com- path, writing or updating messages at Otto Wittner, Centre for Quantifiable Quality plete path-finding problems of large every node visited. Messages are of Service in Communication Systems (Q2S), scale. Given its fully distributed nature, updated according to the evaluated NTNU, Norway it is a promising candidate for path man- cost of the path. E-mail: [email protected] agement in the NGI. The figure illustrates the typical components in a CE-ant After a number of agent life cycles, the system. A high number of ant-like agents overall distribution of messages will traverse the network in search of paths converge toward a pattern indicating not between given source-destination node only the optimal path but also alternative

32 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

ISEE – A Framework for the Evolution and Analysis of Recurrent Neural Networks for Embodied Agents by Martin Hülse, Steffen Wischmann, and Keyan Zahedi

Complex phenomena like bistability, periodic, quasi-periodic and chaotic attractors can already be observed in small artificial recurrent neural networks. How can such a rich reservoir of emergent properties of neural networks be applied to challenging applications?

We utilize the framework of plex systems provided by nonlinear cou- The wheel-driven robot Do:Little is used Evolutionary Robotics (ER) and pled neural systems. for experiments in swarm robotics. We Artificial Life (AL) to evolve recurrent used our evolutionary environment to neural networks (RNNs), which enable The versatility of our proposed evolu- either evolve one controller, which is robot agents to act successfully in an tionary environment is sketched in the then homogeneously distributed within a open and changing environment while figure, where some sample applications group of up to 150 robots, or for evolv- mastering a given task. The combination are shown. A co-evolution strategy was ing in parallel a heterogeneous set of of ER and AL allows the emergence of applied for the ring-shaped robot RNNs, where each individual gets a dif- new control strategies within the sensori- micro.eva, where five moveable arms ferent RNN. In this way, we evolved motor loop, as no a priori knowledge is used to bias a possible solution. Hence, it serves as a source of inspiration for Evolution research into basic neural structures and environment. general principles of neural signal processing systems, which may find applications beyond robotics.

Using a special evolutionary algorithm, the ENS3 (evolution of neural systems by stochastic synthesis) implemented as part of the ISEE package (integrated structural evolution environment, see figure), recurrent neural network structures of the general type for robot control have been developed. The structure and size of RNNs to evolve are open to the evolutionary process, and parameters like synaptic strengths are optimized simultaneously.

The ENS3 can also be utilized to extend inside the ring must maintain a rotation RNNs for solving cooperative tasks, or couple existing RNNs to achieve addi- of the whole body on two passive rollers. such as collecting energy and efficient tional functionality. In this way, different Each arm was considered as an exploration of the environment. behavioural (sub-)functionality can be autonomous agent. Hence, the neural nonlinearly integrated in one network control for each arm was evolved in par- However, the evolutionary environment solving a global robot task. On the one allel but in separated populations. In one is not limited to the optimization of neu- hand, this strategy allows an incremental result, division of labour could be ral controllers. We also successfully evolution of complex control structures observed. The same evolutionary strat- evolved morphological parameters of a solving robot tasks, including increasing egy was successfully applied to evolve bipedal robot and its neural control at the numbers of subtasks and their effective neuro-controllers for the Sony Aibo same time. As a result, we achieved coordination. On the other hand, such robot. Here, three different submodules bipedal walking on flat surface with min- incrementally evolved RNNs give us a for controlling the forelegs, the hind legs imal control and high energy efficiency, wide variety of empirical set-ups with and the coordination of the legs were using its passive dynamics for walking. which to investigate multifunctionality evolved in parallel to fulfil stable and It only needs to compensate the loss of and robust behaviour changes in com- fast walking patterns. energy due to friction and heel strike,

ERCIM News No. 64, January 2006 33 SPECIAL THEME: Emergent Computing

which is solved by a comparably small identify attractors of dynamic systems as function approximation, parameter neural controller. such as RNNs. optimization and so on. The strength of the ISEE package is therefore its versa- Finally, in the figure one can see a snap- With our software architecture it is pos- tility and its powerful analysis tools. shot of some tools that have been inte- sible to evolve different behaviours for grated into the software package. With arbitrary robotic platforms. One can Link: tools providing visualization of RNNs either use fast simulations or even physi- http://www.ais.fraunhofer.de/INDY and their neural activation during robot- cal robots for the evolutionary process, environment interaction, one can gain a as well as for a detailed analysis of con- Please contact: deeper understanding of the relationship trol dynamics during robot-environment Keyan Zahedi, Fraunhofer Institute for between neural dynamics and observed interactions. Although we mainly Autonomous Intelligent Systems – AIS, behaviour. The dynamic properties of applied the ISEE package to robotic Fraunhofer ICT Group, Germany RNNs can also be investigated from a applications, it is not limited to the opti- Tel: +49 2241 14 2253 dynamic systems perspective. For mization of morphology and control. It E-mail: [email protected] instance, one can plot bifurcation dia- can also be applied to various optimiza- grams or isoperiodic plots in order to tion problems unrelated to robotics, such

Emergent Intelligence in Competitive Multi-Agent Systems

by Sander M. Bohte, Han La Poutré

Getting systems with many independent participants to behave is a great challenge. At CWI, the Computational Intelligence and Multi- Agent Games research group applies principles from both the economic field of mechanism design and state-of-the-art machine-learning techniques to develop systems in which ‘proper’ behaviour emerges from the selfish actions of their components. With the rapid transition of the real economy to electronic interactions and markets, applications are numerous: from automatic negotiation of bundles of personalized news, to efficient routing of trucks or targeted advertisement.

In an economic setting, an individual - or For example, we designed a personalized agent - is assumed to behave selfishly: recommendation system in which com- agents compete with each other to peting advertising agents can bid for the acquire the most resources (utility) from attention of the customer. The design of their interactions. In economics, the field the market, combined with adaptive of mechanism design looks at interaction intelligence in the bidding agents, results protocols (mechanisms). Here, the com- in the emergent effect that only the most bined selfish behaviour of individual relevant advertisements are shown to the agents serves a particular purpose as an system user.

emergent property of the system: for Source: dreamstime.com. example, the efficient allocation of Competitive multi-agent games: In a similar vein, we considered the scarce goods. Many computational prob- to each its own. dynamic scheduling of trucking routes lems can be cast as resource allocation and freight. We developed a dynamic problems. Our research transfers the spot market where a software agent in emergent property of human competitive rules should be such that as individual each truck continually participates in multi-agent systems to systems com- agents learn how to optimize their own auctions to increase, change and aug- prised of competing pieces of software, reward, the system as a whole should ment the loads the truck carries. To facil- ie software agents. work with increasing efficiency. Our itate this, we developed bidding strate- work shows that the combination of gies for repeated auctions where the soft- Clearly, software agents in a multi-agent intelligent software agents and well- ware agents compute which combination system must be intelligent and adaptive. designed mechanisms (markets/auc- of loads they can acquire most prof- If intelligent software agents are to work tions/negotiations) can lead to the itably. We show that as an agent tries to out the (local) solutions that are best for desired behaviour of the system as a best anticipate future loads with the aim themselves, the rules of the system must whole, a kind of ‘collective intelligence’. of improving its own profit, an emergent be incentive compatible. That is, the effect is that the market as a whole

34 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

becomes more efficient and the cost of native, more promising bundle composi- schedule should emerge from the agent transport is reduced. tions if the negotiation process stalls. We interactions. designed and implemented a combined Within the same paradigm, we devel- system for making recommendations Obviously, systems where many self- oped methods for dynamically pricing during a negotiation. Extensive simula- interested stakeholders interact pervade information. Electronic information can tions with this system show superior per- our society. Neglecting the strong incen- be sold to many buyers at the same time. formance on a number of benchmarks. tive these stakeholders have to be selfish If we demand that each buyer pays the can be catastrophic for the functioning of same price, the problem is what the price Related to these logistics applications is any system that tries to bring these par- should be. In our system, we show how the use of emergent competitive agent ties together electronically. At the same the selling agent can deduce the pricing systems in health-care logistics. In a pro- time this is also a great opportunity, policy that will maximize revenue from ject on medical information agents, we since we can rely on individual agents to the aggregate of negotiations between have researched the problem of patient maximize their own utility. Thus, we can the seller and the buyers (one-to-many). treatment scheduling in hospitals. design the system to encourage ‘emer- Scheduling complex treatment plans gent intelligence’, as individual intelli- This method has been extended by inte- requires coordination between all the rel- gent behaviour continually improves the grating recommendations in a negotia- evant autonomous departments. Due to workings of the system as a whole. There tion process. A shop aggregates data on the dynamic nature of a hospital, any are many opportunities for smarter soft- customers’ past purchases, and produces approach must be efficient, online and ware agents and better mechanisms, such information on correlations on customer flexible. In cooperation with medical as emergent multi-agent scheduling. We interest in the products on offer. For experts we are investigating the use of are actively pursuing this in current instance, which products are often autonomous software agents negotiating research. bought together? We applied machine- with each other in order to make (or learning techniques to the problem of reschedule) appointments for patient Link: online learning of the bundle combina- treatment. The idea is that as each agent http://www.cwi.nl/sen4 tions that optimize the revenue of a seller follows its owner’s interests and prefer- of information goods. When negotiating ences, agreements over schedules can be Please contact: with a new customer the price of a bun- reached that take into account all the Han La Poutré, CWI dle of such products, the shop uses this individual constraints. Thus, a good Tel: +31 20 592 4082 learned knowledge to recommend alter- E-mail: [email protected]

Building a Computational Digital Economy through Interdisciplinary Research by Petros Kavassalis and Konstantin Popov

Researchers from the Swedish Institute of Computer Science (SICS) and the ATLANTIS Group at the University of Crete are forming a multidisciplinary team to study the complex issues of consumer decision-making in the online world. Their approach combines modern computer modelling and simulation methods with techniques from the emerging field of neuroeconomics.

Consumers on the Web take advantage ation strategies as a way of reducing the tion mechanisms, such as the Online of the fact that information is available in importance of price in consumer choice. Feedback Mechanisms, enable a com- excess. They search and gather product plex choice process and act as positive information, compare prices and check This ‘information excess’ introduces a feedbacks into consumer choice product availability. They access free significant effect: the process of infor- behaviour. reviews in online consumer forums and mation perception and evaluation might share product experience in slashdot-like determine, almost ‘bias’, the entire con- In studying such a complex system, there agoras. Several economists are now talk- sumer decision-making process. are many questions to be answered. ing about a new marketing world in Consumers on the Web will increasingly How are individual preferences being which consumers enjoy a superior infor- have a quasi-realistic experience with constructed and developed? How do mation endowment, and suggest that products, and may make their decision these preferences become coherent in businesses should adopt high-differenti- over the Web alone. In addition, various such an information-rich environment? self-reinforcing, information-propaga- Does the ‘information excess’ increase

ERCIM News No. 64, January 2006 35 SPECIAL THEME: Emergent Computing

or decrease the uncertainty with respect coherently integrates components. can include inferring relationships to the value one should give to various Decision-making will arise from the between elements of data, validation of product attributes? What is the impact of interaction of these cognitive processes, hypothesized models of individual previous consumer experience, if any, on as constrained by the architecture. agents produced by model designers, and preference stability? How efficient can maybe even automated generation and online feedback mechanisms be in mak- Intelligent Agents calibration of models using machine- ing consumers organizationally effective ‘Intelligent’ cyber-consumer agents rea- learning techniques. In this context, anal- and creative under these conditions? The son about products which, in turn, are ysis of data and validation of models of challenge is to develop behavioural represented as vectors of different func- agents become integrated into the agent- models that can help to understand how tional characteristics. This allows for based simulation. Learning of agent consumers make choices in information- product differentiation extending far behaviour through simulation has rich environments, and to investigate beyond typical spatial differentiation, already been proposed by ‘participatory efficient cyber-consumer welfare poli- with uniformly distributed consumer simulation’, where humans substitute for cies and effective business strategies in preferences and products that are spa- model agents during simulation runs. online markets. tially differentiated along a simple unit However, participatory simulation deals circle. Finally, we model the strategies of with humans that entirely substitute for Modelling Digital Economy online marketers to be associated with model agents, whereas in our case, both Algorithmic game theory can provide products and exerted to influence the experimental and publicly available data mechanisms for understanding how competition game. are analysed and continuously used dur- online agents with diverse objectives can ing ‘pure’ agent-based simulation. coordinate adaptively and arrive at a sta- Simulation of these models will generate ble equilibrium. Alternatively, complex highly dynamic, non-linear, experimen- A general-purpose concurrent program- systems methods allow for a greater vari- tal online markets with a high density of ming system like Mozart, can be used as ety of individual behaviour and for sim- circular dependencies and information a platform for developing simulations plicity with respect to the reasoning flows, the study of which can enrich our with cognitive agents and integration of attributed to the individuals. In addition, understanding of real Web markets. Data real-world data. Model agents are con- they allow for individual preferences and from the real Web, such as product list- ceptually concurrent. Their concurrent interaction links between individuals to ings with full descriptive characteristics, implementation can be exploited for dis- be embedded in networks and distributed and product rankings taken from online tributed/parallel execution aiming at across the population with non-uniform forums, should be integrated into the scalability and performance, which probability. system during simulation, making the becomes crucial with large-scale, com- world in which our artificial agents live putationally intensive models. Computer-based modelling and simula- more realistic. Pursuing this line of Integration of cognitive architectures tion can be used to study online markets experimentation, the simulation of a dig- and data analysis into model agents as ‘non-cooperative games’. Neuroeco- ital economy is intended to effectively requires flexibility of a general-purpose nomics can help to investigate the physi- reproduce several structural properties of programming language. cal mechanisms by which the human the real Web, and unfold drivers for the neuroarchitecture accomplishes product Web’s deployment and growth to iden- Links: selection in the online world. These two tify causal (economic) forces at work http://www.c2.org/ areas together can yield very realistic that affect its design. ACT-R: http://act-r.psy.cmu.edu/ behavioural models of ‘intelligent’ Mozart: http://www.mozart-oz.org agents representing cyber-consumers. Integrating Data into Simulation The structure of these agents will be Two types of data can enhance simula- Please contact: more than a framework for implement- tion. The first is information on the pos- Petros Kavassalis ing rational or steady-state adaptive sible behaviour of cyber-consumer University of Crete, Greece behaviour, and will reflect the organiza- agents, collected through neuro-experi- E-mail: [email protected] tion of the mind when consumers make ments with volunteer participants acting decisions. Hence, it is expected to: (i) in quasi-realistic conditions. The second Konstantin Popov possess capabilities that single out paris data collected from the real Web that SICS, Sweden ticular product attributes as meaningful, will be used to populate the local envi- E-mail: [email protected] (ii) take into account advice from other ronment of the individual agents. This consumers, and (iii) feature proactive- environment is, of course, altered by the ness and deploy internal commitment agents’ actions during the course of the ability. Agents’ structure should there- simulation. We propose to use a sim- fore include a number of cognitive ele- ple(r) model of the system as the context ments (sensation, cognitive control, for analysis and exploitation of data. learning, memory, processes for social Whenever state and inputs of a simulated behaviour etc) and a cognitive architec- agent match the available data, its out- ture, such as ACT-R (see Links) that puts should also match the data. Analysis

36 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

Models of Multilateral Cooperative Behaviour by Milan Mareš

The concept of artificial intelligence includes the possibility of cooperation between its representatives (eg robots, complex man-machine systems or other structures), where each is attempting to maximize the values of its goal function. Such an endeavour is formally defined as a cooperative game in which players coordinate their strategies in order to maximize the individual pay-offs for each of them. After a long period of investigation into deterministic games, our attention has recently been focused on games in which the cooperation is distributed among several parallel interests.

The classical theory of cooperative games assumes that, after the negotiation period, the set of players separates into disjoint coalitions and each player participates in exactly one of them. This assumption is not realistic – in real communities with non-trivial structure, regardless of whether their members are human beings or coordinated artificial decision-makers, each member simultaneously follows several goals, distribut- ing its ‘power’ among them. To achieve To illustrate the difference between the these goals, the player cooperates with subtlety of both concepts of fuzzy several groups of partners. Such dis- coalitions, let us consider a simple three- tributed cooperation is usually modelled player game with players a, b, c and by fuzzy coalitions, which have been possible deterministic coalitions (blocks) φ investigated since the early eighties. K0 = , K1 = {a}, K2 = {b}, K3 = {c}, K4 = {a, b}, K5 = {a, c}, The classical model defines fuzzy coali- K6 = {b, c}, K7 = {a, b, c}. tion as a fuzzy subset of the set of all Consider a fuzzy coalition L with players. This approach displays many membership function λ, where λ(a) = 0.3 , advantages, but also has some problems. λ(b) = 0.7, λ(c) = 1.0. Then there exist The fundamental nature of these several fuzzy coalitions of the new (block- becomes evident when the formal model wise) type representing the above is to be compared with its intuitive inter- activities of individual players. For pretation. For example, such a concept of example, M with membership function µ , µ µ µ λ µ λ ν fuzzy coalitions practically excludes where (K!) = (a) = 0.3, (K2) = (b) = 0.7, (K3) = (c) = 1.0; N with memberships , ν ν π π their disjointness (the classical fuzzy set where (K5) = 0.3, (K6) = 0.7; or P with , where (K7) = 0.3, π π theoretical disjointness practically elimi- (K6) = 04, (K3) and some others. Each represents a different structure of the concern of nates the multilaterality of cooperation). players participating in coalition L via particular blocks. The total concern of each player is Consequently, it eliminates also the prin- unchanged even if its distribution among blocks varies. ciple of superadditivity and related game-theoretical concepts specifying the Fuzzy coatlitions. distribution of the coalitional pay-offs.

This recent research was supported by find an adequate model of the complex the fact that players usually divide their grants from the Grant Agency of the structure of the cooperative relations ‘power’ (investments, energy, social Czech Republic No. 402/04/1026 and divided into several fields of interests. contacts, authority etc) among several the Grant Agency of the Academy of interests in which they cooperate with Sciences of CR No. A1075301. Its goal The suggested alternative approach is equivalent (similarly motivated) part- was to suggest an alternative model of based on the definition of a fuzzy coali- ners. These blocks enter the conflict as cooperative behaviour with distributed tion as a fuzzy subclass of the class of all relatively compact units and the total cooperation, better reflecting the super- deterministic coalitions (ie of all subsets incomes of particular players comprise additivity and related properties of fuzzy of the set of all players; in this model, the sum of their incomes from each coalitions. The main motivation was to called ‘blocks’). This approach reflects block. Each block can be active in sev-

ERCIM News No. 64, January 2006 37 SPECIAL THEME: Emergent Computing

eral fuzzy coalitions. In this sense, the the definition of superadditivity, subad- They reflect in some sense similar rela- disjointness concept can be simplified: ditivity and convexity, with interesting tions among players, but offer many the simultaneous participation of a consequences for the concept and prop- more details regarding the subtle differ- player in several fuzzy coalitions (inter- erties of the core and other solutions. ences of their inner structure. mediated via his participation in differ- Another, perhaps even more interesting ent blocks) does not impair their disjoint- advantage is that the alternative model Please contact: ness if he acts in them as a member of offers much more sophisticated tools for Milan Mareš, Institute of Information different blocks. the characterization of the cooperation. It Theory and Automation, Academy of is easy to see that the classical fuzzy Sciences / CRCIM, Czech Republic This model displays certain advantages. coalitions can be transformed into sev- Tel: +420 26605 2803 One of them is the formal simplicity of eral fuzzy coalitions of the new type. E-mail: [email protected]

Emergent Walking Behaviour in an Aibo Robot

by Cecilio Angulo, Ricardo A. Téllez and Diego E. Pardo

A totally distributed neuro-evolved architecture, designed for the general control of autonomous robots, has been employed for the emergence of walking behaviour in the 12 degrees of freedom (DOF) quadruped robotic Aibo dog. The concept of central pattern generators in animals is bio-inspired, and is used to obtain the desired walking robot. Moreover, the search to dissolve the ‘Body-Mind’ problem is our inspiration.

Our project is focused on the reactive Biological CPGs are composed of groups the information from its associated interpretation of the mind, so we are of neurons that produce oscillatory sig- device (ie received sensor information or interested in the emergence of certain nals without oscillatory inputs. It has commands sent to the actuator). Figure 1 behaviour. Quadrupedal walking is a been discovered that the walking move- illustrates a simplified two sensor/two complex task. In a robot like Aibo, there ments of cats and dogs are governed by actuator architecture. are twelve degrees of freedom, and their such elements, and it is thought that coordination to obtain a walking pattern humans behave in the same way. We Through the use of a neuro-evolutionary becomes very difficult. To deal with this implement artificial CPGs using artificial algorithm, modules learn how to cooper- problem, the Knowledge Engineering neural networks (ANNs). To overcome ate and how to control or interpret their Research Group (GREC) at the the problem of capturing the dynamics of associated elements, thereby allowing Technical University of Catalonia (UPC) the system, we have used continuous- the whole robot to learn how to walk. has designed a neuro-evolved controller time recurrent neural networks for the Aibo robot using neural networks (CTRNNs). It should be stated that when several distributed all over the robot. We have IHUs work together on a control task, based our neuro-control system on The proposed distributed neuro-evolved each element has its own particular view Central Pattern Generators (CPGs). architecture is based on several uniform of the situation, since each is in charge of modules known a different sensor or actuator. This leads as Intelligent to a situation where each unit uses its Hardware Units knowledge both of the global situation (IHU). These can and of its particular device to decide be designed what its next action will be. around any physical device in the In this way, the traditional ‘Mind-Body’ robot (sensor or problem can be dissolved. Control of the actuator). Each process is performed through three ele- IHU is composed ments (modellers, controllers and trans- of a sensor or an lators). These elements must adapt the actuator and a relationship between the ‘body’ (embod- Figure 1: ‘Me’ depends on the ‘task’ (task-directed training) as micro-controller iment) and the ‘environment’ (situated- interpreted by the translator; on the ‘environment’ (‘outer world’) that implements ness). interpreted by the modeller; and on the ‘body’ (‘inner world’, an ANN. The sensor and actuator), acting through the controller. ANN processes

38 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

The modeller is a control element that tries to adapt what the sensor ‘sees’ from the ‘outer world’ to what the cerebrum ‘sees’ in its ‘inner world’ by also considering the actions of the controller. Is it the nervous system? In any case, it undergoes continuous learning in order Figure 2: Aibo walking sequence. to adapt the body to the environment.

The translator is a control element that sent to its associated actuator. Is it the to be able to model higher-level tries to translate the external set point (a cerebral neural network? In any case, it behavioural aspects. In this way we behaviour associated to a task) as an must drive the actuator changing the avoid hybrid models that integrate both a interpretation of the ‘outer world’. Is it body/environment situation. ‘deliberative’ and an embodied ‘reac- the behavioural system? In any case, it is tive’ component. the learning function for the whole ‘inner Until now, embodied AI approaches world’ system. have focused on lower-level aspects of Links: the behaviour related to their embodi- http://www.ouroboros.org/evo_gaits.html The controller is a control element that ment; this is the reactive interpretation of http://www.upcnet.es/~upc15838/ deals with both, the internal perception the mind. However, our proposed dis- of the ‘outer world’ (sensor) in the form tributed architecture is able to accom- Please contact: of the ‘inner world’ units (modeller), and plish complex tasks, like the walking Cecilio Angulo-Bahón, Ricardo Agamenón the task to be accomplished (external set behaviour of the 12 DOF Aibo robot Téllez-Lara, Diego Esteban Pardo-Ayala, point) translated to a internal set point, (Figure 2), by dissolving the ‘mind- Technical University of Catalonia (UPC) / also in the ‘inner world’ units (transla- body’ problem. The next stage in this SpaRCIM, Spain tor); then appropriated commands are project is to extend the architecture so as Tel: +34 93 8967798

Concept-Based Text Representations for Categorization Problems by Magnus Sahlgren

Standard practice in most of today’s research on text categorization is to represent texts simply as the bag of words they contain, ignoring both syntax and semantics. As an alternative to this, we have developed a novel form of text representation that we call Bag-of-Concepts, which constitutes a distributed representation of the concepts contained in a text.

Overwhelmingly, the most common rep- lytic methods such as singular value One serious problem with BoC resentational scheme in text categorization decomposition. approaches is that they tend to either be research is the Bag-of-Words (BoW) computationally expensive or require approach. Here a text is represented as a It is important to note that feature extrac- external resources such as dictionaries. vector whose elements are frequency- tion methods handle problems with word To overcome this problem, we have based weights of the words in the text. variability by one of two methods. Either developed an alternative approach for These BoW vectors are then refined, by they group together words that mean producing BoC representations based on feature selection for example, meaning similar things, or they restructure the Random Indexing (see ERCIM News words are removed from the representa- data (ie the number of features) accord- No.50, July 2002). This is a vector space tions according to statistical measures ing to a small number of salient dimen- methodology for producing ‘context such as document frequency, information sions, so that similar words get similar vectors’ for words based on co-occur- gain or mutual information. Another representations. Since these methods do rence data. Very simply, this is achieved refinement method is to use feature extrac- not represent texts merely as collections by first assigning a unique random tion. In this case, ‘artificial’ features are of words, but rather as collections of ‘index vector’ to each context in the data. created from the original ones, either by concepts - whether these be synonym Context vectors are then produced by using clustering methods such as distribu- sets or latent dimensions - we suggest summing the index vectors of the con- tional clustering, or by using factor ana- that a more fitting label for these repre- texts in which words occur. (For an sentations is Bag-of-Concepts (BoC). introduction to random indexing, see

ERCIM News No. 64, January 2006 39 SPECIAL THEME: Emergent Computing

http://www.sics.se/~mange/random_ind sionality much lower than the number of Our experiments also showed that it is exing.html). The point of the context words and contexts in the data, we also always the same categories that are vectors is that they represent the relative achieve a reduction in dimensionality as improved using BoC. This suggests that meanings of words; they can also be compared to the original BoW represen- we might be able to improve the perfor- used to compute the semantic similarity tations. mance of the classifier by combining the of words. two types of representations. When To evaluate the BoC representations, we doing so, the result improves from We use the context vectors produced have used them for text categorization, 82.77% to 83.91% for all categories. For with random indexing to generate BoC which is the task of assigning a text to the top ten categories, the result representations by summing the one or more predefined categories from a improves from 88.74% to 88.99%. (weighted) context vectors of every given set. Our experiments use a support While the difference is admittedly small, word in a text. The resulting BoC vectors vector machine classifier for a standard the increase in performance when com- are effectively combinations of the context categorization collection, and we bining representations is not negligible, cepts (ie word meanings) that occur in have shown that the BoC representations and indicates that concept-based text the text. Note that the representations are outperform BoW with 88.74% vs. representations deserve further study. produced using standard vector addition, 88.09%, counting only the ten largest which means that their dimensionality categories. This suggests that BoC repre- Link: never increases even though the data sentations might be more appropriate to http://www.sics.se/~mange/ might grow: the dimensionality of the use for large-size categories. vectors is a parameter in random index- Please contact: ing. Since we typically choose a dimen- Magnus Sahlgren, SICS, Sweden

A Framework for Efficient Statistical Modelling

by Daniel Gillblad and Anders Holst

Researchers at SICS have developed a new framework for statistical modelling, which can effectively describe complex probability distributions over a large number of attributes of mixed type. This can be very useful when building adaptive models of complex real world data.

In emergent computing it is important In a graph model, a distribution over that real world data are modelled in an many attributes is expressed as a product appropriate way. This means that the of factors, each involving distributions model should represent the relevant over much smaller number of attributes: aspects of reality in a way that is robust, adaptable and data driven. Simple linear regression models are usually not power- By hierarchically combining mixture ful enough to model complex real world models (sums) and graph models (prod- data. We have developed a very general ucts), we have a powerful modelling framework for statistical modelling, framework capable of expressing very Figure 1: A simplified view of the which can effectively describe complex complicated distributions. complementary nature of mixture models probability distributions over a large and graph models. number of attributes of mixed type. Figure 1 shows a simplified view of the complementary nature of mixture mod- The Hierarchical Graph Mixtures els and graph models. A mixture splits Framework the data into subsets of samples, {γ1, γ2} The modelling framework, Hierarchical etc. The graph model instead groups Graph Mixtures, is based on a combina- attributes, eg {x1, x2} and {x2, x3}. By tion of mixture models, probabilistic building hierarchical models, we can graph models, and Bayesian statistics. construct arbitrary groups of attributes and samples. For example, we can use In a mixture model, a complex distribu- different dependency structures within tion is represented as a sum of simpler different clusters of the data. distributions, The meta-structure of the model (ie Figure 2: where to use graphs and where to use The model structure of the sliver detector.

40 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

mixtures) is specified manually. AvestaPolarit). The task was to identify model over one continuous and one dis- Thereafter all parameters of the model which steel coils are at risk of surface crete attribute, which was again realized are trained from data, including graph damage (or ‘slivers’). There were about as a mixture. So in effect we had a mix- structures and mixture partitions. To 270 attributes to consider, both continu- ture model of mixture models of graphs make training and usage of arbitrary hier- ous and discrete. Furthermore, it turned of mixture models over the attributes. An archical structures work, all operations out that different steel types had signifi- overview of the model is shown in on mixtures and graphs are expressed in cantly different sensitivities to slivers. Figure 2. This seemingly complicated such a way as to be independent of the model manages to improve the accuracy form of the sub-distributions. This also In this case we used the hierarchical twenty-fold in identifying which steel provides for a straightforward implemen- graph model at its extreme. We built a coils are at risk of getting slivers. tation of the modelling framework. All mixture with one model for non-sliver parameter estimation is performed using cases and one model for sliver cases, and Link: Bayesian statistics to ensure stability. within each of these we built a mixture http://www.sics.se/iam model over each of eight different steel An Application Example types. Within each of these we modelled Please contact: An example of the application of this the data over the 270 attributes with a Anders Holst, SICS, Sweden model is in the hot steel mill Outokumpu graph model, and finally, in the graph E-mail: [email protected] Stainless AB in Avesta (formerly model we sometimes had to make a joint

Rules, Inferences and Robust Approximation at Work by Antonín Dvořák, Vilém Novák, Viktor Pavliska

How can a computer be taught to act like an experienced human being in complex situations? What sort of human-readable information can be extracted from numerical databases? Can we obtain new robust methods for approximation of functions and for solving differential equations? In addition, can we apply them, for example in signal filtering, compression and fusion of pictures, or the determination of ancient sea levels, when only very imprecise data are available? Soft computing methods are able to provide all of these, and even more.

Soft computing is a scientific discipline in Sets of fuzzy IF-THEN rules represent The computer acts (performs inferences) which methods are developed that work linguistic descriptions of control, deci- similarly to a human such that it resem- effectively in presence of indeterminacy sions and other complex situations. We bles a specific ‘human partner’. We can (ie imprecision and uncertainty), and pro- model these rules using formal fuzzy also model sophisticated human reason- vide practically optimal and inexpensive logic, with a strong emphasis on proper ing, such as that accomplished in detec- solutions in situations where only rough, linguistic treatment of the expressions tive stories or complex decision-making. imprecise information (very often contained within them. This is accom- expressed in natural language) is avail- plished by means of: Linguistic descriptions can be obtained able. • careful study of evaluating linguistic from experts, or by learning from data, expressions (eg ‘small, more or less or by combination of both. For example, Soft computing has been studied at the high, approximately 26, roughly linguistic knowledge (in a human- Institute for Research and Applications medium’, etc). Such expressions are friendly form) can be extracted from of Fuzzy Modelling (IRAFM) at the indispensable in human reasoning databases for complex queries contain- University of Ostrava in the Czech • interpreting fuzzy IF-THEN rules as ing vague notions. Another possibility is Republic for several years. In particular, linguistically expressed logical impli- to build a linguistic description from a two concepts have been investigated both cations (the theoretical framework is successful course of control, eg of a tech- theoretically and experimentally: fuzzy higher-order fuzzy logic) nological process, mobile robot etc. IF-THEN rules and fuzzy approximation. • using a special inference method called Using our methods, we can also search Recall that the former take a form such as Perception-based logical deduction. the so-called linguistic associations in “IF salary is more or less small AND numerical data, eg “a significantly small quality of house is medium THEN mort- The methodology enables us to commu- crime rate AND a large proportion of gage coefficient is rather small”. nicate with the computer in (restricted) residential land IMPLY more or less natural language, without needing to medium housing value”. Such associa- penetrate into the fuzzy logic machinery. tions characterize relations in the data in

ERCIM News No. 64, January 2006 41 SPECIAL THEME: Emergent Computing

a way that is closer to the way of think- tion of fuzzy IF-THEN rules and devel- sion and/or fusion of pictures. Our meth- ing of experts from various fields. oped a new method called the fuzzy (F- ods are very robust, that is, they have a )transform. This is a powerful method low sensitivity to changes in the input Fuzzy approximation is a class of meth- for approximating functions, which has a data (eg signal filtering depends very lit- ods for the approximation of classical wide variety of applications, eg in signal tle on sampling). functions using techniques of soft com- processing, approximate solutions of dif- puting. We have elaborated approxima- ferential equations (ordinary as well as Our software system LFLC (Linguistic tion methods using relational interpreta- partial), or in methods for the compres- Fuzzy Logic Controller) deals with linguistic descriptions and enables fuzzy Figure 1: Simulink approximation; an interface to MAT- interface of LFLC - LAB/Simulink is also available (see simulation of Figure 1). A large-scale application of control of simple our methods can be found in dynamic system. Kovohut\v{e} B\v{r}idli\v{c}n\’a, in the Czech Republic, where LFLC con- trols five massive aluminium furnaces.

IRAFM is a partner of the Research Centre DAR (Data/Algorithms/Decision Making), headed by the Institute of Information Theory and Automation of the Czech Academy of Sciences (an ERCIM member). Our goals in DAR include fuzzy modelling of complex pro- Figure 2: cesses (where linguistic descriptions Demonstration of play a key role) and a combination of filtering abilities of stochastic and fuzzy models. F-transform. Link: http://irafm.osu.cz/irafm

Please contact: Antonín Dvo ák University of Ostrava, Czech Republic Tel: +420 597 460 218 Particle Swarm Optimization for the Reconstruction of Permittivity Range Profiles from Microwave Measurements

by Simone Genovesi and Emanuele Salerno

At the Signal and Images lab, ISTI-CNR, we are developing a new algorithm to reconstruct the permittivity range profile of a layered medium from microwave backscattering data. The algorithm is based on a particle swarm strategy to optimize a specific edge-preserving objective functional. Our technique is able to efficiently find the global optimum of the objective functional, while preserving the discontinuities in the reconstructed profile.

Inverse scattering is relevant to a very and industrial engineering, non-destruc- sensitive to noise or produce over- large class of problems, where the tive testing, medical imaging and subsur- smoothed profiles by enforcing global unknown structure of a scattering object face inspection. However, a truly satis- constraints. This is a major drawback, is estimated by measuring the scattered factory solution to the inverse scattering since the discontinuities carry essential field produced by known probing waves. problem has not so far been found owing information on possible heterogeneous As illustrated in Figure 1, several appli- to its intrinsic ill-posedness. Common inclusions in the building material, and cation areas are involved, such as civil inversion techniques are either highly maintaining them in the reconstructed

42 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

profile is very important. found by the swarm, and is able Moreover, the nonlinear relation- to evaluate the status of its cur- ship between the scattered field rent location. The initialization of and the object function, and the the position and velocity of each robustness of the inversion algo- agent is random. We are now rithms, are still open issues and running some preliminary simu- most scattering tomography tech- lations to test different swarm niques are not sufficiently reli- sizes and other initialization pro- able to solve practical problems. cedures, and to tune the parameters involved in the velocity Our current purpose is to recon- update (eg particle inertia, and struct one-dimensional permittiv- social and cognitive rate). ity range profiles of architectural objects from microwave Our first results have been backscattering data on a certain Figure 1: Inverse scattering has a wide range of encouraging. Figure 3 shows a sim- frequency range. The complete applications. ulated profile with large disconti- iterative procedure involves a nuities, and its reconstruction from forward solver and an optimizer. noisy data as obtained by our The former computes the backscattered Our unknowns are the permittivity val- swarm algorithm. The wall was assumed field from the currently proposed solu- ues of all the layers and the locations of to be 24cm thick and was subdivided tion. The latter uses the results from the possible discontinuities. To optimize the into twelve layers. Note that the permit- solver plus some a priori information to objective functional, we are implement- tivity values are large (this structure drive the optimization process. We have ing a particle swarm algorithm. Particle could model a concrete wall with an chosen to discretize the wall into a finite swarm optimization (PSO) is an evolu- internal air inclusion). This means that number of homogeneous and lossless tionary computation technique inspired our test object is strongly scattering, and layers of equal thickness (see Figure 2). by the social behaviour of flocks of birds many other inversion techniques would It is assumed that the total thickness of and swarms of insects. In PSO, each have failed to reconstruct it. the wall is known and the incidence of solution is represented by an agent, or the probing waves is normal. particle, which explores the multidimen- This research is conducted within the sional solution space. framework of a project financed jointly In order to estimate the wall permittivity by the Italian National Research Council as a function of the depth coordinate z, During the search procedure, the agents (CNR) and the Italian Ministry of we build a functional containing two dis- change their positions over time by fly- Education and Research. tinct terms. The first is a suitable dis- ing around in the solution space. Since tance between the measured and the cal- we adopt a fully connected swarm topol- Please contact: culated backscattering data. The second ogy, each agent knows the best location Simone Genovesi, ISTI-CNR, Italy is a combination of a quadratic, first- found by the rest of the swarm, and is E-mail: [email protected] order, smoothness constraint and an able to adjust its velocity according to its explicit smoothness-breaking term, own experience and the experience of all which preserves possible abrupt permit- the other agents. As a result, any agent is tivity variations where these are likely to stochastically attracted towards both its occur. own best location and the best location

Figure 2: Interaction between the multilayered structure and the microwave radiation.

Figure 3: Results from a simulated discontinuous profile (25 dB- SNR data).

ERCIM News No. 64, January 2006 43 SPECIAL THEME: Emergent Computing

Building Blocks from Biology for the Design of Algorithms for the Management of Modern Dynamic Networks

by Gianni A. Di Caro, Frederick Ducatelle, Luca Maria Gambardella, and Andrea Rizzoli

Modern computer and communication networks are becoming increasingly large, heterogeneous and dynamic. Traditional network algorithms fail to deal efficiently with this increased complexity. The EU-funded project BISON addresses this problem by drawing inspiration from biology to provide the building blocks for a new family of distributed, self-organizing, adaptive and scalable network algorithms.

Biology-Inspired techniques for Self- behaviour at the system level as a result Organization in dynamic Networks of local interactions and self-organiza- (BISON) is a three-year Shared-Cost tion. Examples of systems showing this RTD Project (IST-2001-38923) funded highly complex behaviour are ant by the Future and Emerging Technologies colonies, and the cells of the human activity of the Information Society immune system. Because of these Technologies Program of the European appealing properties, natural systems Commission. It runs from January 2003 have served as a source of inspiration until December 2005. The BISON con- for a number of successful algorithms sortium draws on multidisciplinary exper- and frameworks, mainly for optimiza- tise from the University of Bologna tion tasks (eg evolutionary computation From Biology to Dynamic Networks. (Italy), which is the coordinator, Dalle and ant colony optimization). The Molle Institute for Artificial Intelligence application of the biological approach (IDSIA) (Switzerland), the Technical to network problems has attracted rela- (MANETs), which are MWAH networks University of Dresden (Germany) and tively little attention. However, the anal- in which the nodes are mobile and can join Telenor AS (Norway), which is the indus- ogy between networks and natural sys- or leave the network at any time, and sen- trial partner. tems is evident: the nodes of a network sor networks, in which nodes are usually environment can be seen as the units of a not mobile and have specific sensing capa- The objective of BISON is to develop biological system, and the actions and bilities. In the case of MANETs, we new network algorithms that are adap- input/output of users form the external addressed the issue of the optimization of tive to changes, robust to failures and environment. the routing function to allow the network perturbations, work in a self-organized to cope effectively with the problems and decentralized way, and are able to The BISON research intends to exploit raised by mobility and interference. For function efficiently in heterogeneous this analogy. We investigate biological this purpose we developed AntHocNet, a large-scale systems. Central to BISON’s processes in a systematic way, and traffic- and topology-adaptive algorithm approach is the idea of drawing inspira- abstract and reverse-engineer the basic with both reactive and proactive compo- tion from biological systems to develop mechanisms at work. This allows us to nents. AntHocNet is designed after ant this new family of algorithms. The ratio- identify building blocks for the design of colonies and their ability to find the short- nale behind this choice comes from the distributed and self-organizing network est paths in distributed and dynamic envi- following observations. Biological sys- algorithms displaying adaptivity, robust- ronments by using a combination of tems usually have the ability to effec- ness and scalability. BISON has focused repeated and concurrent path sampling, tively adapt to constantly changing envi- on a number of specific network envi- pheromone laying/following (stigmergic ronments. Moreover, they are usually ronments and core functions. Peer-to- communication), and stochastic decisions. robust to internal perturbations or loss of peer (P2P) and multi-hop wireless ad hoc units, and are able to survive and evolve (MWAH) networks are the dynamic For sensor networks we focused on the in a wide range of environments. environments being investigated. The problem of the distributed assignment of functions are routing, topology manage- node transmission ranges. Here the aim Biological systems have obtained these ment, content search, monitoring, data was to establish topologies that can pro- properties through evolution: they are aggregation and load balancing. vide full connectivity while minimizing usually composed of a large number of energy consumption, thereby maximiz- dynamic, autonomous and distributed At IDSIA, we have focused on two net- ing network lifetime. For sensor net- units, which display effective adaptive works. These are mobile ad hoc networks works we developed both an exact cen-

44 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

tralized approach and an effective dis- are behind T-Man, a protocol provably rithms for routing and topology control in tributed heuristic based on a reaction-dif- able to build a variety of complex struc- wireless networks.) In other cases, such fusion model, which is characteristic of a tured topologies. The pattern-matching as the protocols for aggregate calcula- number of biological processes (eg mor- and cell proliferation characteristics of tions, excellent empirical performance is phogenetic development). Still in the antibodies of the human immune system also accompanied by pleasing theoretical context of ad hoc networks, but wired have provided basic inspiration for an properties of convergence. ones, we also developed a distributed algorithm that effectively performs con- active monitoring system based on the tent searching in unstructured overlays. Links: same ant colony behavior at the heart of Diffusion processes and chemotaxis, a BISON web site: AntHocNet. process describing cell movement in http://www.cs.unibo.it/bison response to concentration gradients of IDSIA web site: http://www.idsia.ch In P2P networks, all nodes or users can chemicals, are at the foundations of a sys- Ant Colony Optimization: realize bidirectional and symmetric com- tem for load balancing in distributed stor- http://iridia.ulb.ac.be/~mdorigo/ACO/ACO.html munications. An overlay connecting the age networks. A new proactive protocol Artificial Immune Systems: peers is established above the IP layer, for the calculation of aggregate functions http://www.dca.fee.unicamp.br/~lnunes/ which provides the underlying communi- (eg average load) has been derived from immune.html cation functionalities. Important issues in the patterns common to the epidemic IETF MANET page: http://www.ietf.org/ P2P networks include strategies for per- spreading of contagious diseases. html.charters/manet-charter.html forming general network-wide calcula- Sensor networks: http://www.research. tions and searches, as well as building In conclusion, the observation of biologi- rutgers.edu/~mini/sensornetworks.html and maintaining overlay topologies. cal processes has provided numerous P2P networks: http://p2p.internet2.edu/ These can facilitate critical operations ideas for the design of a number of fully like content search/publishing and join- distributed, adaptive, robust and scalable Please contact: ing/leaving. In BISON we successfully network algorithms. The experimental Gianni A. Di Caro, Frederick Ducatelle, addressed most of these issues. results from extensive simulations pro- Luca Maria Gambardella, Andrea Rizzoli vide a strong validation of the approaches Istituto Dalle Molle di Studi sull’Intelligenza Cell replication mechanisms provided followed. Performance is usually good, Artificiale (IDSIA), Manno-Lugano, Switzerland inspiration for a protocol that builds and and in some cases the proposed algo- Tel: +41 58 6666660 maintains random overlay topologies rithms clearly outperform state-of-the-art E-mail: over time, while ideas from cell adhesion algorithms. (This is the case of the algo- {gianni,frederick,andrea,luca}@idsia.ch

Chemical Programming of Self-Organizing Systems by Jean-Pierre Banâtre, Pascal Fradet and Yann Radenac

Chemical programming relies on the ‘chemical metaphor’: data are seen as molecules and computations as chemical reactions. This programming paradigm exhibits self-organizing properties and allows the description of autonomic systems in an elegant way.

The Chemical Programming Model reached; that is to say, when no more reac- Chemical Programming This formalism was proposed to capture tions can take place. For example, the com- and Self-Organizing Systems the intuition of computation as the global putation of the maximum element of a non- Autonomic computing provides a vision evolution of a freely interacting collection empty set can be described as: in which systems manage themselves of atomic values. It can be introduced replace x, y by x if x ≥ y according to some predefined goals. The through the chemical reaction metaphor. essence of autonomic computing is self- The unique data structure is the multiset (a Any couple of elements x and y of the mul- organization. Like biological systems, set possibly containing identical elements), tiset is replaced by x if the condition is ful- autonomic systems maintain and adjust which can be seen as a chemical solution. filled. This process continues until only their operation in the face of changing A simple program consists of a reaction the maximum element remains. Note that components, workloads, demands and condition and an action. Execution pro- in this definition, nothing is said about the external conditions, such as hardware or ceeds by replacing elements that satisfy the order of evaluation of the comparisons. If software failures, either innocent or mali- reaction condition with the elements speci- several disjoint pairs of elements satisfy cious. The autonomic system might con- fied by the action. The result of such a pro- the condition, reactions can be performed tinually monitor its own use and check gram is obtained when a stable state is in parallel. for component upgrades. We believe that

ERCIM News No. 64, January 2006 45 SPECIAL THEME: Emergent Computing

chemical programming is well suited to Adding new ill-ordered values breaks the of files. The editors are distributed over a the description of autonomic systems. It ‘equilibrium’, since the ordering property network and each works on a different captures the intuition of a collection of is violated. However, sort searches conversion, making modifications to the files cooperative components that evolve tinuously for new ill-ordered values and and committing them locally. From time freely according to some predefined con- causes reactions so that the state will to time, two or more editors merge their straints (reaction rules). System self- reach a new stable state. modifications to propagate them. This sys- management arises as a result of interac- tem can be easily described with reaction tions between members, in the same way This very simple example shows how rules to reflect the self-organization of the as ‘intelligence’ emerges from cooper- chemical programming can naturally system. The system is also self-repairing: ation in colonies of biological agents. express self-organizing systems. A pro- if an editor loses his local version it can gram is made of a collection of rules revert to a previous state by synchronizing A Self-Organizing (active molecules), which react until a with one or several other editors. Sorting Algorithm stable state is reached and the corre- Consider the general problem of a system sponding invariant properties satisfied. Future Plans whose state must satisfy a number of These rules remain present and are We are currently studying the application properties but which is submitted to applied (without any external interven- of this model to the coordination of pro- external and uncontrolled changes. This tion) as soon as the solution becomes gram execution on grids. In a first step, system must constantly reorganize itself unstable again. applications are programmed in an to satisfy the properties. Let us illustrate abstract manner, essentially describing this class of problem by a simple sorting Examples of Applications the chemical coordination between (not example where the system state is made An autonomic mail system has been necessarily chemical) software compo- of pairs (index : value) and the property described within the chemical frame- nents. In a second step, chemical service of interest is that values are well ordered work. Rules ensure that all messages programs are specifically provided to the (ie a smaller index means a smaller have reached their destination; if it is not run-time system in order to obtain the value). If the environment keeps adding the case some reactions occur to reach expected quality of service in terms of random pairs to the state, the system must that state. The system includes rules that efficiency, reliability, security etc. reorganize itself after each insertion of an ensure other autonomic properties such ill-ordered element. The system is repre- as self-healing (rules that set and unset Link: sented by a chemical solution “State = emergency servers in case of failures), J.-P. Banâtre, P. Fradet and Y. Radenac. ”, consisting of self-optimization (rules that balance the Higher-Order Chemical Programming Style. pairs and the following active molecule: load between several servers), self-pro- In Proceedings of the Workshop on sort = replace (i : x), (j : y) tection (rules that suppress spam or Unconventional Programming Paradigms by (i : y), (j : x) if i < j and x > y viruses), self-configuration (rules that (UPP’04), LNCS 3566. Springer-Verlag. forward messages to the new address of Also available as preprint at: The molecule sort looks for couples of a user) and so forth. http://www.irisa.fr/paris/Biblio/Papers/ ill-ordered values and swaps them. The Banatre/BanFraRad04UPP.pdf solution evolves up to the point where no We have also specified a chemical more reactions are possible: the solution Distributed Versioning System. In this Please contact : has reached a stable state and the order- application, several editors may concur- Jean-Pierre Banâtre, Pascal Fradet, ing property is satisfied. rently edit a document consisting of a set Yann Radenac, INRIA/IRISA, France E-mail: {Jean-Pierre.Banatre, Pascal.Fradet, A Development Model for Robust Fault-Tolerant Design

by Andy Tyrrell and Hong Sun

Multi-cellular organisms, products of long-term biological evolution, demonstrate strong principles for the design of complex systems. Their nascent processes, including growth, cloning (self-replication) and heeling (self-repair and fault- tolerance), are attracting increasing interest from electronic engineers. All of these characteristics are encoded in the information stored in the genome of the fertilized cell (zygote). The process of growth from a single zygote to a mature organism is called development.

Development is controlled by genes, interactions between different proteins, The development of an embryo is deter- which determine the synthesis of proteins. between proteins and genes within cells, mined by these interactions. Figure 1 The activity of genes sets up the complex and hence the interactions between cells. shows a representation of the pattern for-

46 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

exhibit the same structure. Each The model is defined using the VHDL cell has direct access to its six language, with Xilinx Virtex XCV1000 neighbouring cells. No direct as its target FPGA. The evolutionary access to non-adjacent cells is process is designed to be implemented in allowed in the current model. hardware, and can therefore be consid- Figure 2 shows the connections ered Intrinsic Evolvable Hardware between cells. The control unit (IEHW).

(CU) in Figure 2 generates the state and chemical of the cell. Using a process of evolution and devel- The state decides what type the opmental growth a French flag pattern cell will be, while the chemical can be formed using this hardware. (It Figure 1: French flag model of pattern formation. constructs the development envi- should be noted that functional circuits ronment. The Execution Unit have also been produced using this same (EU) would perform any func- technique.) An interesting emergent mation of cells controlled by the concentra- tional activities required by the cell (for property of all of these systems is their tions of a morphogen. Based on the level of example logical functions required as capacity to recover from different faults. morphogen in each cell, the cells develop part of any arithmetic operations). into different patterns according to thresh- For example, we can change all red cells old values. This cellular model can be used for the into blue, white cells are changed into task of pattern formation (the shape of grey (dead state) and blue cells are Inspired by natural evolution, Evolvable different flags or other shapes) or func- switched to red. Hence all the states of Hardware (EHW) was developed in the tional circuits (multiplier or even parity, cells are wrong and the chemicals are all last decade as a new approach to the etc). The difference between pattern for- set to 0. In five steps the system recov- design of electronic systems. It has mation and functional circuits is that dur- ered from this fault, and maintained the demonstrated the ability to perform a ing pattern formation, the fitness of the correct shape thereafter. wide range of tasks from pattern recogni- system is based on whether the states of tion to adaptive control. A variant of this the cells are in accord with the expected In another example, all cells apart from approach mimics the developmental pro- states, the EU not being required. When the top left cell are set to dead and their cess from embryo to multi-cellular organ- developing functional circuits, the state chemicals are set to 0. A correct shape ism, constructing robust and fault-tolerant generated by the CU is used to control appears in four steps, although this turns circuits. the EU function and the fitness of the into a shape with a fault in another two system is based on the output of the EU. steps, and finally regains the correct Motivated by the importance of the Here we illustrate an experiment looking shape in another two steps. interactions in and between cells in at the pattern formation of the French biology, this work seeks an approach to Flag, so only the CU is used. The honeycomb model presented here is increase these interactions, inves- part of a larger research project, but this tigating whether this benefits evo- simple example shows its capacity for lution and development. pattern formation: its ability to develop Honeycomb structures, often pro- into a specific shape, maintain that shape duced by natural evolution and and recover to the shape after many claimed by architects and kinds of transient faults are injected. economists to be the most efficient structure for covering an Basic cell processes in embryonic devel- area, are used. The concept of a opment are imitated. A morphogen is morphogen is employed when used to mimic the pattern formation pro- judging the state of a cell. The cess of the embryo. Chemicals are used proposed model mimics various to build up the environment for develop- cell developmental processes. ment. The interactions between cells are Although some processes (eg cell based on the transmission of chemical movement) are restricted by the and state information. fixed hardware architecture, those such as changes in cell state, cell- Link: to-cell signalling, and cell death http://www.elec.york.ac.uk/intsys/ are modelled. Please contact: The cells used in this cellular Andy M. Tyrrell, University of York, York, UK model are homogenous; they all E-mail: [email protected] interpret the same genome and Figure 2: Connections between cells.

ERCIM News No. 64, January 2006 47 SPECIAL THEME: Emergent Computing

Emergent Properties of the Human Immune Response to HIV Infection: Results from Multi-Agent Computer Simulations

by Ashley Callaghan

Results obtained from multi-agent computer simulations of the immune response to HIV infection suggest that the emergence of certain properties, or lack thereof, may play a critical role in determining the relative success of the response.

The normal immune response to infec- will generally lack the ability to respond incorporate many of the key entities that tion by a virus or bacterium (antigen) is to new strains of the virus. participate in the immune response to characterized by a complex web of inter- HIV infection. The agent-based models actions involving numerous cells and The Need for Models of the we employ for our simulations model molecules of the host’s defence system. Immune Response to HIV each cell as a unique entity with a set of The response results in the emergence of The question now arises as to why we characteristics that, together with clearly clones of cells and molecules that would wish to model this system? The specified rules, define its functional together manage to remove the antigen main reason we need models of the behaviour. This approach allows us to from the body. However, the situation immune response to HIV is that a lot of incorporate a number of important char- with regards to infection by HIV is dif- experiments that biologists would like to acteristics, including cell-surface recep- ferent. While the immune system gener- carry out in order to test hypotheses can- tors, physical location in the system, ally manages to bring the viral load (ie not be performed. The reasons for this affinity for a particular antigen, and so the level of HIV) to low levels within include ethical considerations, and the on. The use of advanced techniques such as parallel or distributed computing means we have now reached a stage where it is possible to design simulations incorporating cell population levels that are fast approaching the level found in the human body.

Emergent Properties during the Immune Response To investigate possible causes that may explain why different individuals show such a wide range of responses to HIV abinfection, we performed numerous simulations in which different ‘patients’ were Figure 1: Graphs of viral dynamics and immune response for three hypothetical infected with the same quantity of an patients during the acute phase of infection: (a) viral load per mL of plasma; (b) number identical strain of the virus. By different of activated T Killer cells (CD8) per mL of plasma. patients, we mean different runs of the simulation, with the only variable being weeks to months of initial infection, HIV fact that biological experiments are very the random seed used to determine the is never completely eliminated but often incredibly expensive and time con- characteristics of the various cells that instead remains present in low concen- suming. If suitable computer or mathe- together make up the immune system. In trations. The reason HIV manages to matical models existed which could order to demonstrate the critical role that escape the immune response is most faithfully reproduce what is known from the emergence of a particular property likely due to two factors – its incredible the literature, then it may be possible to plays in the relative success of the rate of replication and high rate of muta- carry out certain experiments ‘in silico’ response, results for three of these hypo- tion. New strains of HIV constantly (ie computer-based simulations), instead thetical patients are discussed. emerge during infection as a result of the of the more usual in vitro and in vivo error-prone nature of replication and the experiments. Figure 1a illustrates that the simulations rapid turnover of virus in infected indi- quite accurately capture the viral load viduals. These strains have the potential Multi-Agent Computer Simulations dynamics associated with the acute stage for immune escape, as the clones of cells With the computational resources avail- of infection as described previously. that develop against the wild-type virus able to us today, we are in a position to That is, a rapid rise in viral load is fol- construct computational models that lowed by a sharp decline to what is often

48 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

referred to as a ‘set point’, resulting from response that emerges in the case of Further Work the emergence of T Killer cells and HIV patient 3 is very poor, with the result that As noted previously, one of the most neutralizing antibodies. the viral load at the set point is signifi- striking features of HIV is its very high cantly higher than in the other two cases. rate of mutation. Throughout the course Figure 1b shows the emergence of acti- It has previously been shown that the of the infection, mutant strains of the vated T Killer cells. These cells attempt higher the viral load at the set point, the virus are constantly emerging. The rela- to destroy cells infected with HIV before worse the long-term prognosis with tive success of these mutant strains they have a chance to release new virions regards the progression of the patient to across different patients is a subject that into the peripheral blood. Comparison of AIDS. This therefore suggests that the warrants future investigation.Please con- these two figures suggests that the emer- emergence of a sufficient T Killer tact: gence of a sufficiently large T Killer response during the early stages of infec- response is critical in bringing viral load tion may play a critical role in determin- Ashley Callaghan, to a low level at the set point. In contrast ing the length of time before the patient Dublin City University / IUA, Ireland with patients 1 and 2, the T Killer develops full-blown AIDS. E-mail: [email protected]

Network Emergence in Immune System Shape by Heather Ruskin, and John Burns

Individual experiences of disease can be extremely diverse. As a result, if targeted treatment of the individual is to be achieved, we need some way of assessing the variables that lead to different outcomes for different people. Visualizing the record of viral attacks in terms of a network of connections that are formed and broken in defence against an attack, can aid in understanding observed effects. This is similar to mapping and redrawing operational outcomes on a larger scale, in order to incorporate necessary or unforeseen changes. Here we describe some of the ideas involved in the formation of connected networks of information at the cellular level, where these provide immune system fingerprints of individual exposure.

Emergent principles of the immune sys- (cell-types) as in Table 1 – exchanges Emergent behaviour is thus explicitly tem T-cell repertoire and its self-organi- information with the shape space, which exhibited. zation can be simply modelled for a lym- models affinity between cell types and phatic compartment, using stochastic clearance pressure. The CTL population The focus here is on two cell- types only, cellular automata. The immune system grows exponentially in response to APC APC and CTL. We omit discussion of hybrid model incorporates a shape space stimulation, with a rate dependent on the the humoral response as well as refer- formalism, permitting each activated distance between the APC and CTL in ence to CD4+ T-cells – their role in facil- effector T-cell clonotype and viral epi- shape space. The model permits: itating CTL activity is implicitly tope to be represented as nodes of a • faithful replication of real-space recir- assumed. Dynamics, governing the graph, with edges modelling the affinity culation dynamics affinity between antigen and lympho- or clearance pressure applied to the anti- • representation of T-cell repertoire cyte, arise from the shape space formal- gen-presenting cell (APC,) which bears clonotype distribution and its differen- ism (see Perelson A.S. and Oster G.F. the target epitope. For repeated exposure tiation, antigenic drift, antigenic shift (1979), Theoretical Studies of Clonal to infection by heterologous or mutating ,immune memory and cross-reactivity Selection: Minimal Antibody Repertoire viruses, a distinct topology of the net- • the evolution of the shape space to be Size and Reliability of Self-Non-Self work space emerges. Properties of the studied through complex network theory. Discrimination. J. Theor. Biol., 81 emergent network reflect recent experi- (4):645-70). Each unique antigen epi- mental results on cytotoxic T-cell (CTL) tope and CTL clone is represented as a activation, apoptosis, cross-reactivity point within the two-dimensional space and memory, especially with respect to of given size. Surrounding each CTL re-infection. clone is a disc of radius r. Any antigen epitope located within the disc is sub- Biological systems are complex by jected to a clearance pressure with a nature, and the immune system is typical force inversely proportional to the sepa- in this respect. In the hybrid model out- ration distance (see Burns J. and Ruskin lined, the real space of the lymphatic Table 1: Notation and definition of model H.J. (2004) Diversity Emergence and compartment – with entities and states entity states Dynamics during Primary Immune

ERCIM News No. 64, January 2006 49 SPECIAL THEME: Emergent Computing

Figure 1: Development of 4- homologous viruses, with cross-reactive epitope network. Only CTL epitopes. clonotypes are shown. Centres and edges = immunogenic Early and protective immunity can be epitopes and stimulated CTL mediated by memory T-cells generated respectively. By (c) there is a pool by previous heterologous infection (rep- of memory CTL clonotypes, some resented graphically by nodes with close enough in shape space to degree > 2). Cluster linkage illustrates exert clearance pressure against conditions by which immunity to one the heterologous virus, (non-leaf virus can reduce the effects of challenge nodes deg (2)). By (d), by another (see Figure 1). Damage to or suppression of final infection sees suppression of critical cross-reactive ‘a’- cross-reactive contribution of nodes has significantly greater impact memory cells, specific to a virus, than damage to leaf or ‘b’-nodes. thus resulting in a network Different disease outcomes to identical connection. infection strains can be explained in terms of density and distribution of a- nodes. The degree of protection from Response: A Shape Space, Physical memory pool consumes a given percent- cross-reacting memory cells depends on Space Model: Theor. in Biosci., age of activated CTL, so nodes remain the distance between the memory T-cell 123(2):183-194). A network model of active in shape space, preserving edge clonotype and immunogenic epitope, shape space emerges naturally from the connections to stimulatory epitopes. with optimal immunity for re-infection real-space model, with each immuno- by the same antigenic epitope. Results genic epitope ek and activated CTL Recent work indicates that effector CTL show increasingly effective clearance clonotype cj represented by a node in the memory cells can recognise epitopes of dynamics as the memory pool increases space. Clearance pressure applied unrelated viruses, so that heterologous and each T-cell clone has a finite chance between two nodes forms a directed edge viruses (derived from a separate genetic of becoming a long-lived memory cell. between them, with a weight dependent source) may be a key factor in influenc- on the affinity. After initial infection, ing the hierarchy of CD8+ T-cell Please contact: most cj undergo programmed apoptosis responses and the shape of memory T- Heather J. Ruskin (a crucial regulator of immune system cell pools. Shape space can be used to Dublin City University / IUA, Ireland homeostasis), which corresponds to edge model both homogeneous viruses with E-mail: [email protected] deletion. However, recruitment to the conserved and mutated epitopes, and

Agent-Based Modelling of Viral Infection

by Dimitri Perrin

The three phases of the macroscopic evolution of the HIV infection are well known, but it is still difficult to understand how the cellular-level interactions come together to create this characteristic pattern and, in particular, why there are such differences in individual responses. An ‘agent-based’ approach is chosen as a means of inferring high-level behaviour from a small set of interaction rules at the cellular level. Here the emphasis is on cell mobility and viral mutations.

One of the most characteristic aspects of ence with respect to HIV infection is so of the individual, together with the vari- the HIV infection is its evolution: in the diverse. In particular, the work aims to ous factors that influence this. If such initial short acute phase the original viral address questions relating to variation in ‘priming patterns’ can be recognized or strains are destroyed, in the second year- the length of the individual latency even predicted, then in the long term we long latency period, the number of period. This may be very long (for rela- may have a way of ‘typing’ an individual strains slowly increases and, in the final tively low success of antipathetic muta- and targeting intervention appropriately. phase, Acquired ImmunoDeficiency tion) in one individual, compared to Unfortunately, understanding how the Syndrome (AIDS) develops when the another with much higher mutation lev- immune system is primed by experience immune system is no longer able to cope els. of antigenic invasion and diversity is with the multiplying strains and is over- non-trivial. The challenge is to deter- come. The principal aim of this work, The indications are that the observed mine what assumptions can be made based at Dublin City University, is to try variation lies in the priming and initial about the nature of the experience, and to understand why the range of experi- level of fitness of the immune response then modelled, tested against clinical

50 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing data and hence argued plausibly. The The lymph node (adapted from aim is to understand how the cell interac- N. Levy, Pathology of lymph tions lead to the observed endpoints. nodes, 1996) is modelled as a What exactly is involved in antigenic matrix in which each element is diversity? How variable is the mutation a physical neighbourhood and rate and the viral load? What is the can contain several agents of importance of cell mobility and how each type. realistic is this in terms of cross-infection and subsystem involvement? How important then is the cross-reactivity? lymph node (as a test case). This proto- use of these methods is a natural conse- The immune response is dynamic and type, however, includes all known inter- quence and advantage of the multi-agent includes growth and replenishment of actions contributing to cell-mediated approach. A human body contains hun- cells and in-built adaptability, through immunity and the local evolution of the dreds of lymph nodes. The aim here is to mutation of its defences to meet new virions. The antibody-mediated response extend the size and complexity of the threats. It also includes aspects of cell has not been considered initially, systems that can be modelled to some- mobility, which may be captured by because the cell-mediated arm plays a thing approaching reality. means of defining the movement and dominant role in repelling attack. The affinity of cell-types in a defined spatial agents implemented represent Th The representation of the innate response framework. In particular, this will enable (helper) and Tc (cytotoxic) lymphocytes, as a common background and the adap- us to study the variation in viral load and Antigen Presenting Cells and virions. tive part as a characterized set of features the way in which the host response may They inherit from a common C++ class will be the next step. This will allow the lead to degradation of protection. designed to deal with features such as development of a large system to be mobility. Each class then implements studied over a longer period, in order to To investigate these questions, an through attributes and methods the spe- focus on disease progression endpoints ‘agent-based’ approach is chosen as a cific properties of each cell type, such as and intervention effects. means of inferring high-level behaviour the activation of a Tc cell by a Th cell. from a small set of interaction rules at the The lymph node is modelled as a matrix The author would like to thank the Irish cellular level. Such behaviour cannot be in which each element is a physical Research Council for Science, extracted analytically from the set of neighbourhood able to contain various Engineering and Technology for the rules, but emerges as a result of stochas- agents of each type. funding made available through the tic events, which play an important part Embark Initiative. in the immune response. The next step is to increase of the number of lymph nodes. This extension Please contact: The initial model consists of agents (or involves millions of agents and requires Dimitri Perrin, Dublin City University/ functional units) with designated proper- major computational effort, so that paral- IUA, Ireland ties that mimic the operation of a single lelization methods are inevitable. The E-mail: [email protected]

The ‘Decent’ Project: Decentralized Metaheuristics by Enrique Alba and Martin Middendorf

The project ‘Decent’ (Decentralized Metaheuristics) is developing new swarm- based metaheuristics in which a decentralized design leads to emergent phenomena. These are not only important for solving complex problems, but are suitable for parallel execution in computing grids and multi-task hardware platforms.

Metaheuristics, such as evolutionary ered to be the state-of-the-art methods; have been proposed for most meta- algorithms (EA), ant colony optimiza- these areas include engineering applica- heuristics. However, most approaches do tion (ACO), simulated annealing (SA) tions, bioinformatics, telecommunica- not utilize the full potential of parallel and particle swarm optimization (PSO), tions, logistics and business. execution because of their synchronicity are methods that can be applied success- and execution on clusters of homoge- fully to almost all optimization prob- Due to their practical importance and the neous machines. All this makes it diffi- lems. For many hard tasks in a huge vari- need to solve large real-world instances cult to apply them to interesting parallel ety of areas, these algorithms are consid- of hard problems, parallel algorithms systems such as dynamically reconfig-

ERCIM News No. 64, January 2006 51 SPECIAL THEME: Emergent Computing

Figure 1: Test function (left) and emergent behaviour of the algorithm; swarm particles after 0, 50 and 400 iterations.

urable hardware (eg FPGAs) and large (DNA fragment assembly, protein struc- defined in the population, eg having five heterogeneous networks (computational ture etc). individuals). grids). In the first year of the project, a decen- Figure 2 illustrates the evolution of a In the ‘Decent’ project, strictly decentralized clustering algorithm that can grid of individuals that separately evolve tralized versions of metaheuristics that cluster data packets in networks was to solve the same problem. do not suffer from the restrictions of designed and applied to Particle Swarm Neighbourhoods interact by means of standard parallel/distributed metaheuris- Optimization (PSO). PSO is a meta- implicit information exchanges achieved tics are developed. ‘Decent’ is a two- heuristic inspired by bird flocking by the application of recombination of year joint project between researchers behaviour. A PSO algorithm maintains a their shared individuals. Stochastic from the University of Málaga (Spain) swarm of particles, which move through mutation of the individual contents is and the University of Leipzig the search space in the search for an opti- also performed, and new individuals are (Germany). The project is funded by mal value. The movement of a particle is kept at each grid position if they are bet- MECD (Ministerio de Educación y influenced by its velocity and the posi- ter than the existing ones. Such decen- Ciencia, Spain) and DAAD (German tions where good solutions have already tralized behaviour means that diversity is Academic Exchange Service) within the been found, either by the particle itself or preserved in the solution space. It also ‘Spain-Germany Integrated Actions’ by other particles in the swarm. Figure 1 allows a graceful convergence to an opti- framework (HA2004-0008). The main shows a test function that is to be minimal solution that would be hard or even focus of the project is on decentralized mized. It can be seen that the clustering impossible to locate with other algo- algorithms that are suitable for dynami- helps the PSO to explore different areas rithms. The work in progress includes cally changing heterogeneous networks, containing minimum function values. self-organization of the emergent as mobile ad hoc (and sensor) networks behaviour and parallelization in clusters and dynamically reconfigurable comput- Another example of emergent behaviour and grids. ing systems. One main goal is to investi- appears in cellular Genetic Algorithms gate the emergent properties of such (cGA). This is a kind of population- Most of this work is related to the decentralized swarm-based algorithms, based technique in which tentative solu- Spanish TRACER project and to the not found in the separate behaviour of tions are evolved on a given topological European EMBIO project. their components. This is expected to structure, eg a toroidal 2D mesh (see have a significant impact both in the Figure 2). Each individual in such a pop- Links: field of advanced algorithms and in ulation has a fitness (quality of a prob- http://neo.lcc.uma.es applications. This is particularly the case lem solution) that is iteratively improved EMBIO project http://www-embio.ch.cam.ac.uk for complex problems arising in by applying operators to the set formed TRACER project: http://tracer.lcc.uma.es telecommunications (routing, coding, by one individual and its neighbours broadcasting etc) and bioinformatics (much in the manner of traditional GAs, Please contact: but on overlapped small neighbourhoods Enrique Alba University of Málaga / SpaRCIM, Spain Tel: +34 952 132803 E-mail: [email protected]

Martin Middendorf University of Leipzig, Germany Figure 2: Snapshots for a cellular Genetic Algorithm (cGA) every fifty iterations until a Tel: +49 341 9732275 problem solution is located (black point in rightmost figure). The brighter an individual E-mail: [email protected] is coloured the higher is its fitness (black means a solution has been found); green individuals are within 80% of the optimum value and red ones are within 90%.

52 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

Evolutionary Methods make New Effective Laser Shapes by Thomas Bäck and Joost N. Kok

Controlling the behaviour of atoms and molecules by very short laser pulses using so-called femtosecond lasers is a very active and challenging research field. In a collaboration between the Leiden Institute of Advanced Computer Science (LIACS) and the Institute for Atomic and Molecular Physics, Amsterdam (AMOLF), evolutionary algorithms are used to optimize the shape of such laser pulses, based on a reformulation of the task as a high-dimensional, nonlinear optimization problem. These optimization methods, gleaned from the model of organic evolution, turn out to be well suited for tackling this challenging task.

Traditionally, the advancement of phys- In problems in physics that depend on a Figure 1: The instrumental set-up for the ical understanding through experimental large number of parameters, great laser pulse shaping experiment. The research involves the definition of con- advances can be made using a new mask is controlled by a large number of trolled experiments in which a problem approach based on evolutionary algo- continuous parameters, which affect the of interest is studied as a function of one rithms. The large number of parameters final form of the pulse in a highly or more relevant experimental parame- limits the usefulness of experiments nonlinear way. ters. The outcome of the experiment where only some of these parameters are then provides insight into the specific varied in a prescribed manner. An evolu- role of these parameters. This approach tionary approach is a viable alternative in The key advantage of this iterative opti- dates back to the days of Galileo. In a many of these situations. In this mization approach is that one does not famous series of experiments, he mea- approach the system of interest is studied need to know a priori the details of the sured how far a ball rolls down a gradi- within a closed loop strategy, where in working mechanism of the complex sys- ent as a function of the parameter time, each iteration the set of system parame- tem. Instead, the goal is to learn about and concluded that the distance travelled ters is modified to some extent by means the underlying physics by interpreting by the ball is proportional to the square of specialized mutation and recombina- the sets of parameters produced by the of the time. This approach has led to an tion operators. After doing an actual evolutionary algorithm. This is in con- enormous wealth of accumulated know- experiment on the system with these trast to performing experiments with ledge. However, it fails when the num- parameters, the best performing values controlled variations (ie knowledge- ber of parameters relevant to the prob- for achieving a given objective (also based or trial-and-error-based variations lem of interest becomes very large. called fitness in evolutionary algorithms) by human experts) of these parameters. These days more and more of these situ- are selected for the next round. Because of the generic nature of the evo- ations are encountered. lutionary approach, this methodology can be applied to a wide variety of different situations.

For as long as efficient optical sources have been available, scientists have tried to use optical means to control the behaviour of atoms and molecules. Specifically, with the availability of easy-to-use tunable lasers, numerous efforts have been undertaken to control the dynamics (dissociation, ionization, reactivity) of chemically and biologically relevant species. In a joint Dutch ‘Foundation for Fundamental Research on Matter’ (FOM) project between Figure 2: Three different pulse shapes which have been obtained by evolutionary LIACS (Professors Bäck and Kok) and algorithms (left), and the corresponding course of the optimization process plotted as AMOLF in Amsterdam (Professors quality of the pulses over time (right). It is clear that, due to the niching technique Vrakking, Herek, Muller and Tans), employed, very different pulse shapes have been found which are nevertheless very interesting results have been obtained in similar in their final quality, thus confirming the high complexity of the underlying the field of femtosecond laser pulse optimization problem. shaping using evolutionary algorithms.

ERCIM News No. 64, January 2006 53 SPECIAL THEME: Emergent Computing

There is currently great interest in the can be steered, and optimal laser pulse Physics can be, and clearly demonstrate atomic and molecular physics commu- shapes for a given optimization target advances in both fields (namely, opti- nity in aligning molecules with laser can be found. The target function is the mized pulse shapes and new concepts in pulses, since dealing with an aligned alignment of an ensemble of molecules evolutionary algorithms). Moreover, as sample of molecules simplifies the inter- after interaction with a shaped laser the project started just one year ago, a pretation of experimental data. To con- pulse. Using so-called niching methods, variety of additional results over the trol the motion of atoms or molecules by which make sure that an evolutionary course of this collaboration are expected. irradiating them with laser light, one has algorithm yields several alternative solu- to provide laser pulses with durations on tions, new effective laser pulse shapes Link: the same time scale as the motion of the were detected. http://www.liacs.nl/research/Algorithms/ particles. The outline of an experimental setup for such an experiment is illus- Recent results (see Figure 2 for pulse Please contact: trated in Figure 1. By applying a self- shapes and the corresponding course of Thomas Bäck, LIACS, Leiden University, learning loop using an evolutionary evolution) have shown how fruitful the The Netherlands algorithm, the interaction between the cooperation between researchers from Tel: +31 71 527 7108 system under study and the laser field Evolutionary Algorithms and Molecular E-mail: [email protected]

Neural Net Modelling in Financial Engineering of Options

by Jerome Healy, Longchuan Xu, Maurice Dixon, Fang Fang Cai, Brian Read and Brian Eales

Financial options play a key role in the risk management carried out by financial institutions. As a consequence, their pricing and hedging against adverse market movement is a major challenge for loss avoidance. Although standard options are regulated by the exchanges, exotic options are traded over-the-counter. Neural networks provide a powerful way of deriving models from traded data that are free from most financial modelling assumptions.

A standard call (put) option gives the establish a Financial Modelling Research holder the right but not the duty to buy Base at Moorgate in the City. (sell) an asset on or before a future date at a price specified now. This enables There is now an abundance of high- and the holder, for a premium, to hedge low-frequency data available from vari- against adverse asset price changes ous exchanges. Analytic models have Figure 1: Prediction Bands for Synthetic while benefiting from favourable move- tended to assume an underlying log-nor- Example — y(x) is the true regression for ments. mal pricing distribution following a benchmark synthetic model; d repre- Black-Scholes-Merton. Numerical mod- sents the target data points generated by The Financial Market Modelling project els free from this and other constraints adding a known noise function to it. d* is aims at developing a computational can be extracted from traded data. the estimate of the target. L and U are the framework with techniques appropriate true upper and lower prediction intervals to accurately predicting option prices The aim of the project is to provide a sys- while L* and U* are the approximate pre- from exchange data, and to hedging their tematic framework for the generation of diction intervals obtained from the predic- movement against adverse movements. numerical models and predictions/fore- tion for the variance (squared error). This would enable issuers of options to casts from option market data. This will buy and sell financial instruments to be done by combining established mod- change their exposure as the market elling approaches with diagnostic tech- moves. niques from econometrics.

The research is grounded in the underly- The development framework was entiable; also they are used to provide ing data-mining program of the Business CRISP-DM (CRoss Industry Standard self-error estimates. Linear regression and Information Technology Department Process for Data Mining). It was special- and Monte Carlo simulations supported of CCLRC. Recently a project grant was ized to option market applications. the investigation. Standard commercial obtained from HEFCE by London Neural nets were deployed to provide database, spreadsheet, and data-brushing Metropolitan University, Londonmet, to regression models that are twice differ- tools were linked to a data-mining work-

54 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

bench, SPSS Clementine and Figure 2 SEI Put Option RNDs - One-Month to the econometric tool eViews Forecast of FTSE 100 Level. One-month from Quantitative Micro forecasts of the FTSE 100 closing price from Software. the extension of RNDs to SEI (American) put options that can be exercised early. The Initial work concentrated on forecast values (red line) are the medians of demonstrating high accuracy the RNDs. The heavy black dashed line is the (R2>0.99), but avoiding over- FTSE 100 closing price one month (17/18 fitting, using neural nets for trading days) later. The dotted green and blue pricing standard options with lines are ±2 standard deviation (≈95.46%) fixed expiry dates from daily confidence intervals. The true values lie within data. These had been traded as the confidence band in all cases. ESX options on the London International Financial Futures and Options Exchange, LIFFE. An important advance was the develop- early exercise put options, SEI, is shown cation in the recent ERCIM-managed ment of a practical method for obtaining, in Figure 2. project, TELEMAC. The visualization for unknown targets, confidence of pre- approaches applied in TELEMAC have diction intervals. It uses the capacity of a Future work will include over-the- been shown to be very helpful for the neural net to model the joint probability counter (OTC) options, which are priced financial data inspection. distribution of a function value and point on the basis of a direct contract between squared error. The method is robust to option buyer and seller. These are not Link: variable error bars. Its application to a exchange regulated and the prices are not http://epubs.cclrc.ac.uk/ synthetic benchmark problem is shown publicly available. In addition, some - search text with authors: “Lambert Dixon” in Figure 1, where there is good signal OTC options have a discontinuity where - search text with authors: “Healy Dixon” recovery from very noisy data. Since the value changes to/from zero if the second derivatives of neural nets can be asset touches a value called the barrier. Please contact: determined, it is possible to obtain an Barrier options pose specific problems Maurice Dixon, LondonMet, UK important probability distribution, the as hedging depends on the value of the E-mail: [email protected] risk neutral density (RND). This means gradients close to a discontinuity. that that forward expectation of an asset Simon Lambert, CCLRC, UK price can be used to aid in setting a mea- The approaches developed here have Tel: +44 1235 445716 sure of value-at-risk. The extension to been applied to an environmental appli- E-mail: [email protected]

Self-Optimization in a Next-Generation Urban Traffic Control Environment by Raymond Cunningham, Jim Dowling, Anthony Harrington, Vinny Reynolds, René Meier and Vinny Cahill

The Urban Traffic Control Next Generation (UTC-NG) project tackles the problem of global optimization in next-generation UTC systems by using sensor data to drive a fully decentralized optimization algorithm.

Current approaches to Urban Traffic Recent advances in sensor technology dination model for traffic-light collabo- Control (UTC) typically take hierarchical have made online vehicle and traffic ration, and are typically reliant on previ- and/or centralized approaches to the flow detection possible. This in turn has ously specified models of the environ- global optimization of traffic flow in an enabled adaptive traffic control systems, ment that require domain expertise to urban environment, often with limited capable of online generation and imple- construct. These models are typically success. By exploiting the increasing mentation of signal-timing parameters. used as an input to both sensor data inter- amounts of available sensor data (eg Adaptive control systems are widely pretation and strategy evaluation, and from inductive loops, traffic cameras, on- deployed throughout the world. may often be too generic to adequately board GPS systems etc), next-generation reflect highly dynamic local conditions. UTC system designers possess a unique However, the adaptive traffic-control opportunity to address the problem of systems that are currently deployed are These systems have a limited rate of adap- global optimization of traffic flows. hampered by the lack of an explicit coor- tivity and are designed to respond to grad-

ERCIM News No. 64, January 2006 55 SPECIAL THEME: Emergent Computing

ual rather than rapid changes in traffic stream junctions. These can then utilize attempt to solve its problem locally by conditions. They employ centralized or this information when choosing the devising an appropriate policy. hierarchical data processing and control appropriate action to take. By operating in However, as the solution to the problem algorithms that do not reflect the localized this completely decentralized way, the depends on local traffic conditions that nature of fluctuations in traffic flow. UTC system obviously becomes self vary over time, the traffic light must con- managing and can, less obviously, be tinually attempt to estimate/learn the Collaborating Traffic Lights designed to optimize the global flow of optimal policy for the junction under its An alternative approach to the one pur- vehicles through the system. The tech- control. sued by existing UTC systems is to allow nique used to achieve this decentralized the controller/agent of the set of traffic optimization through coordination/col- Status lights at a junction to act autonomously, laboration is called Collaborative In a similar approach, UTC-CRL is tak- deciding on the appropriate phase for the Reinforcement Learning (CRL). ing an experimental approach to validat- junction. The actions available to such ing the appropriateness of CRL in a an agent are similar to the those available Distributed Optimization large-scale UTC setting. In particular, an to a traffic manager in a centralized/hier- CRL is a decentralized approach to objective of the UTC-CRL experimental archical UTC system (ie remaining in the establishing and maintaining system- approach is to verify that a consensus current phase or changing to another wide properties in distributed systems. can emerge between collaborating traffic available phase). CRL extends Reinforcement Learning light agents, and that this consensus (RL) by allowing individual agents to allows optimal traffic flow in the large- In a similar manner to existing central- interact with neighbouring agents by scale setting. The envisioned setting for ized/hierarchical UTC systems, the agent exchanging information related to the this work corresponds to the Dublin city would monitor the level of congestion at particular system-wide optimization area, which consists of 248 traffic light the junction under its control based on problem being solved. The goal of CRL junctions, over 750 non-traffic light available sensor data and use this infor- is to enable agents to produce collective junctions, and over 3000 links between mation to decide which action to take. behaviour that establishes and maintains these junctions. Over time, the agent learns the appropri- the desired system-wide property. ate action to take given the current level The UTC-NG project is supported by the of congestion. However, if the agent at a Optimizing the global flow of traffic in a TRIP project, a multi-disciplinary junction simply optimizes its behaviour UTC system can be considered as a sin- research centre funded under the using only local congestion information gle system-wide problem. This can be Programme for Research in Third-Level at that junction, this may result in locally decomposed into a collection of discrete Institutions (PRTLI), administered by optimal performance but also in subopti- optimization problems, one at each traf- the Higher Education Authority. mal overall system performance. fic light junction in the UTC system. Please contact: In order to achieve optimal system-wide Since each traffic light agent has a sub- Raymond Cunningham, Trinity College performance, the set of agents at traffic problem that is unique to that agent, a Dublin / IUA, Ireland light junctions in the UTC system should traffic light agent cannot delegate the Tel: +353 1 608 2666 communicate their current status to agents solution of this problem to one of its E-mail: [email protected] at neighbouring upstream and down- neighbours. Rather, the agent must

Self-Organized Routing in Mobile Ad Hoc Networks using SAMPLE

by Jim Dowling and Stefan Weber

In the SAMPLE project, we are investigating decentralized collaborative learning techniques to develop a routing protocol for Mobile Ad Hoc Networks, where routing agents collectively learn to exploit stable routing paths in the network environment. This approach to routing lends itself to large-scale ubiquitous computing scenarios, in which large numbers of ubiquitous mobile devices are intermixed with static infrastructure networks.

Mobile Ad Hoc Networks (MANETs) niques based on strong consensus and ments. Current approaches to routing in are a promising area of application for global knowledge are breaking down these environments see individual emergent computing techniques. due to the decentralization and MANETs as separate from existing Traditional distributed-systems tech- dynamism inherent in these environ- infrastructure, and view the network

56 ERCIM News No. 64, January 2006 SPECIAL THEME: Emergent Computing

Figure 1: Components of a wireless node and the installation of these nodes in the WAND testbed.

nodes as homogeneous. However, wire- Positive and negative feedback are the different types of traffic in MANETs, less infrastructures are increasingly con- key mechanisms that adapt an agent’s such as data and voice traffic. The devel- sidered to be a community service (also behaviour to both its neighbours and a opment of a protocol to carry voice traf- called muni-wireless networks) that changing environment. In SAMPLE, fic over muni-wireless networks would should be provided by local authorities. routing agents use CRL to explore their allow the provision of free voice calls, Mobile devices in these environments local environment by executing routing something currently only available on exhibit various levels of processing actions, and provide one another with the Internet. However, to achieve this power, mobility and connectivity, but feedback on the state of routes and net- goal, self-organized routing techniques existing approaches do not consider work links. Using CRL, agents adapt will be required for dynamic networks these characteristics. In the SAMPLE their behaviour in a changing environ- where topology, resources and node project, we are developing a routing pro- ment to meet system optimization goals availability are subject to frequent and tocol for metropolitan area MANETs. In while using only local state information. unpredictable change. these networks, mobile nodes collectively learn and self-organize to exploit SAMPLE has been implemented in the Our research is conducted as part of the any fixed or temporary network infra- NS-2 simulator, and the results have CARMEN project, which is funded until structure in the environment. been very encouraging. Simulations 2006 by the Higher Education Authority show how feedback in the selection of (HEA) of Ireland. The SAMPLE project is concerned with links by routing agents enables them to developing a MANET routing protocol self-organize in varying network condi- Links: for WiFi-enabled computing devices tions and properties, resulting in the opti- SAMPLE website: such as laptops, smart mobile phones mization of network throughput. In http://www.dsg.cs.tcd.ie/sites/SAMPLE.html and handheld devices. In MANET rout- experiments, emergent properties such WAND website: ing protocols, computing devices, also as traffic flows that exploit stable routes http://www.dsg.cs.tcd.ie/sites/WAND.html called nodes, act as both consumers and and re-route around areas of wireless Please contact: providers of the routing services in the interference or congestion have been Jim Dowling network. In particular, we have been demonstrated. As such, SAMPLE is an E-mail: [email protected] investigating how nodes can collectively example of a complex adaptive dis- determine the stability and quality of net- tributed system. Stefan Weber, work links in the system. This ability is Trinity College Dublin, Ireland particularly beneficial in the presence of We are now moving to the deployment Tel: +353 1 608 8423 a wireless infrastructure, as nodes learn phase of the project, in which the proto- E-mail: [email protected] to route over the higher-performance sta- col will be tested on a real-world testbed. ble paths in the infrastructure to access The Wireless Area Network for Dublin popular services such as e-mail and the (WAND), which we have deployed in Web. the centre of Dublin city, is a testbed infrastructure covering a 1.5km route The design of the SAMPLE protocol is from Trinity College to Christchurch based on collaborative reinforcement Cathedral. It allows experimentation on learning (CRL), an unsupervised learn- both protocols and applications for ing algorithm that enables groups of MANETs in a metropolitan area. An ini- reinforcement learning agents to solve tial implementation of the protocol has system online optimization problems. been developed for Linux and experi- CRL provides feedback models that map ments will establish its performance in a changes in an agent’s environment and real-world wireless network. its neighbours onto internal changes in the agent’s policy, using distributed As part of the project, we are also inves- model-based reinforcement learning. tigating the use of SAMPLE for carrying

ERCIM News No. 64, January 2006 57 R&D AND TECHNOLOGY TRANSFER

Articles in this Section XSMapper: a Service-Oriented Utility 58 XSMapper: a Service-Oriented Utility for XML Schema Transformation by Manuel Llavador and José H. for XML Schema Transformation Canós, Technical University of Valencia/SpaRCIM, Spain by Manuel Llavador and José H. Canós

59 Analysis and Modelling of Genomic Data A typical case of low-level interoperability, particularly frequent in the Digital by Anna Tonazzini, Francesco Bonchi, Libraries world, is the federation of collections via metadata conversion. Roughly Stefania Gnesi, Ercan Kuruoglu and speaking, a federation consists of a number of repositories, each with its own Sergio Bottini, ISTI-CNR, Italy format, which agree on a common format for metadata exchange. Any metadata 61 Numerical Mathematics Consortium record, must then be transformed into the common format before it is sent as the 62 CORRECT – Developing Fault- result of a request. In this note, we report on a solution to the metadata conversion Tolerant Distributed Systems problem based on semantic mappings. Although it was developed to federate by Alfredo Capozucca, Barbara Gallina, collections in a specific project, XSMapper is domain-independent and can be Nicolas Guelfi, Patrizio Pelliccione and Alexander Romanovsky, University of used in any context where an XML schema transformation is required. Luxembourg Making distributed and heterogeneous is to use XSL transformations between 63 The WearIT@work Project: Empowering the Mobile Worker with systems interoperate has been a chal- records. For a collection to be added to Wearable Computing lenge for researchers and practitioners the BibShare Federation, the owner of by Elisa Basolu, Massimo Busuoli, over the last decade. The complexity of the collection must create an XSL tem- ENEA - Italian National Agency for the problem has led to solutions with plate that transforms the records to the New Technologies, Energy and the Environment and Mike Lawo, increasing levels of sophistication, BBF. However, writing an XSL template University of Bremen, Germany depending on the requirements imposed is not a trivial task, and any tool support- by the domains of the application. ing template generation would represent 64 PULSERS Delivers on Phase 1 – Europe to Adopt a Ruling for Ultra Different forms of middleware represent a significant improvement to the federa- Wide Band the most general solution for achieving tion process. by Walter Hirt, IBM Research GmbH, full interoperability, but in some cases Switzerland simpler solutions can be used. This is Since the problem of document transfor- 66 AVISPA: Automated Validation of particularly the case when the require- mation goes beyond the scope of Internet Security Protocols and ment for interoperability originates from Bibshare, we developed a general solu- Applications the heterogeneity of (meta)data formats, tion to the problem. In its most general by Alessandro Armando, Università di Genova, Italy, David Basin, Luca as often happens in the Digital Libraries version, this be stated as follows: given Viganò, ETH Zurich, Switzerland, world. two XML Schemas S1 and S2 that repre- Jorge Cuellar, Siemens and Michael sent respectively the source and target Rusinowitch, Laboratoire Lorrain de Such a problem arose during the devel- formats of a transformation, obtain as Recherche en Informatique et ses Applications, France opment of BibShare, an environment for automatically as possible the XSL tem- bibliography management, a project plate that transforms S1-valid documents 67 Prague Texture Segmentation Data funded by Microsoft Research into S2-valid documents. Generator and Benchmark by Stanislav Mikeš and Michal Haindl, Cambridge, that allows users to collect Institute of Information Theory and bibliographic references, insert citations XML Semantic Mapper (XSMapper) Automation, Academy of into documents and automatically gener- solves the problem based on the defini- Sciences/CRCIM, Czech Republic ate a document’s bibliography. Unlike tion of semantic mappings between 68 Visiting Hybrid Museums: A Colony former tools, BibShare works with a source and target schemas, following of Ants in your Pocket variety of word-processing systems, and three steps (see figure): by Javier Jaén and José A. Mocholí, permits references to be inserted not only Universidad Politécnica de Valencia / SpaRCIM, Spain from personal citation collections, but 1.Extraction of the concepts that are also from bibliography servers available used both in source and target 70 Edutainment and Game Theory Realization through an Open-Source on the Internet, such as DBLP . As might schemas. A concept is a term used to UNESCO Virtual Perspolis Project be expected, each collection has its own name different elements in an XML by Sepideh Chakaveh, Olaf Geuer, metadata format(s). In order to unify the document. For instance, the concept Fraunhofer Institute for Media- result sets of federated searches and ‘author’ is used to denote the elements Communication, Germany; and Stefan return these data to the user, each record representing the authors of books and Werning, Massachusettes Institute of Technology, USA retrieved must be converted to a com- articles; this means that there will be mon format. We call this the Bibshare different elements ‘book/author’ and 71 CONFIOUS: Conference Management System with Bibliographic Format (BBF). ‘article/author’, which may be trans- Intelligence, Power and Style lated to the same element in a target by Manos Papagelis and Dimitris Given that XML is used to exchange schema (eg following the LaTeX Plexousakis, ICS-FORTH, Greece data, the natural solution to the problem model, an element ‘bibitem/author’).

58 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

This step is performed automatically resulting XML tree (composed of ele- by the XPathInferer Web service. This ments and their attributes), instantiat- service not only finds the concepts, but ing the target schema. The latter also their location within documents in inserts the source schema values in the the form of XPATH expressions. This resulting XML text following the is very important because location in semantic mappings defined in step 2. the document can be a key property during the conversion process. Notice that most of the components of 2.Definition of the semantic mappings XSMapper are available as XML Web between the elements of S1 and S2. services, and can be used at the URLs This step cannot be performed auto- listed below. We are working on a vari- matically, unless some ontology relat- ety of improvements to the tool, with ing the concepts in both schemas can special emphasis on looking for ways to be used to infer them. XSMapper pro- automate the definition of the semantic vides a friendly user interface for mappings that would make XML con- defining three kinds of mappings, version a fully automated task. namely direct, function-based and constant. Direct mappings are used to Links: link one or more concepts of the XSL template generation workflow. Bibshare: http://www.bibshare.org source schema to one or more con- XSMapper: cepts of the target schema that are http://bibshare.dsic.upv.es/XSMapper.exe semantically equivalent (eg the vided by XPath and XSLT to define XPathInferer Web service: ‘author’ presented above). Function- our function-based semantic map- http://bibshare.org/XPathInferer/XPIWS.asmx based mappings are defined in cases pings. Finally, the constant mappings XSLGenerator Web service: where it may be necessary to apply are used when we want to assign a http://bibshare.org/XSLGenerator/ some functions to the source concepts constant value to a target concept. XSLGeneratorWS.asmx in order to get the equivalent target 3.Generation of the XSL template. This elements (for instance, splitting one task is performed automatically by the Please contact: concept like ‘author’ into two concepts XSLGenerator Web service. An XSL José H. Canós, ‘first name’ and ‘surname’). As we are template has two kinds of elements: Technical University of Valencia / SpaRCIM using XSLT to transform documents, structural elements and value-selec- E-mail: [email protected] we can use the set of functions pro- tion elements. The former build the http://www.dsic.upv.es/~jhcanos

Analysis and Modelling of Genomic Data by Anna Tonazzini, Francesco Bonchi, Stefania Gnesi, Ercan Kuruoglu and Sergio Bottini

At ISTI-CNR, Pisa, researchers from different areas of computer science are studying an integrated and interdisciplinary approach to various problems in Computational Biology and Bioinformatics.

The achievements of the Human increasingly crucial in order to be able Researchers from the ISTI Laboratories Genome Project and the rapid develop- to extract useful knowledge from this for Signal and Images, Knowledge ment of post-genomic technology have data. Discovery and Delivery and Formal dramatically increased the importance Methods and Tools form an interdisci- of genomics and genetics in disease Bioinformatics exploits modern comput- plinary group whose comprehensive analysis and diagnosis, novel drug and ing and communication technology, and research in a number of areas of bioin- therapy discovery, and early detection is stimulating research that addresses formatics has been recently formalized or even prediction of disease. The aim computationally demanding challenges in a Work Package of the national CNR is to improve healthcare strategies and, in biology and medicine. This highly project on ‘Computational Biology’. ultimately, the quality of life of the indi- interdisciplinary field includes data min- vidual. Due to the enormous flow of ing, modelling of complex systems, 2D Our main goal is the development of heterogeneous biological data that is and 3D visualization, signal and image models and analysis methods that can being made available, powerful tools processing and analysis, ‘in silico’ mod- help to describe and understand the spa- for storage and retrieval, processing, elling and simulation, and algorithms for tial characteristics of DNA with a func- analysis and modelling are becoming large-scale combinatorial problems. tional value, and the computational

ERCIM News No. 64, January 2006 59 R&D AND TECHNOLOGY TRANSFER

mechanisms behind complex biological gies such as constraint-based mining, systems such as gene regulatory net- and possibly used as input to sophisti- works. A bottom-up strategy will be cated clustering techniques. The ultimate adopted, in which low-level processing goal is to provide simulations and mod- integrates with high-level classification elling of molecular interactions and and modelling. The focus will be on the metabolic pathways. In this respect, we structural analysis of genomes and pro- are also studying formal methods that teins, and on the detection and functional can be used to describe complex biologi- analysis of clusters of genes related to cal systems and verify their properties. underlying biological processes in Due to the real and massive parallelism microarray experiments. involved in molecular interactions, investigations into the exploitation of A large number of genomes, ranging from biomolecular models as examples of viral and microbial pathogens to higher global and parallel computing are also in organisms, have now been fully progress. sequenced and made publicly available

for investigations at various levels. by courtesy of Paolo Gualtieri, Istituto di Biofisica CNR, Pisa The research activity described above is Nevertheless, although DNA sequencing Figure 1: 3D structure of a photoreceptor carried out in collaboration with other is a mature technique and many research protein of Euglena gracilis. institutions in the fields of biomedicine efforts to further improve the algorithmic and informatics. The biomedical institu- phase are reported in the literature, accu- multiple experimental conditions. Large tions provide us with data and validate rate identification of bases has not yet amounts of data are becoming available, the biological significance of the results. been fully achieved by the software of providing simultaneous measurements Our main collaborations are with the available automatic sequencing of expression profiles and of interactions Institute of Biophysics, CNR, Pisa, the machines. In this respect, we are currently of thousands of genes. The challenge is National Institute for Applied Sciences studying unsupervised statistical tech- to discover the complex functional (INSA) in Lyon and the niques to model electrophoresis signals dependencies underlying these data and Immunohematology Unit, II Pisa and correct the colour cross-talk and to identify biological processes driving Hospital Cisanello. We intend to estab- peak-spreading phenomena. At the gene coregulation. At ISTI, techniques lish new collaborations with other bioin- genome scale, we have developed effi- for unsupervised clustering of gene formatics groups, and in particular we cient algorithms for fragment assembly expression maps from microarray data are seeking fruitful interactions within by partial overlap in the shotgun sequenc- are now being investigated. In particular, ERCIM. ing method. As per high-level processing, we are studying statistical techniques of we are working on comparative genomics Blind Source Separation, such as Link: for the identification of conserved and Independent Component Analysis http://www.isti.cnr.it/ResearchUnits/Labs/ invariant structural elements with func- –(ICA), nonlinear and constrained ICA, si-lab/ComputationalBiologyGroup.html tional value within the genomes. Special and Dependent Component Analysis, attention is being paid to the large portion which should provide non-mutually Please contact: of non-coding regions. exclusive gene clusters. The results of Anna Tonazzini, ISTI-CNR, Italy these analyses will be compared with Tel: +39 050 3153136 In proteomics, we take advanced tech- those of local pattern discovery strate- E-mail: [email protected] niques for mining complex and high- dimensional information spaces, and Figure 2: apply them to frequent local pattern dis- Computational covery in protein databases, and to the Biology at ISTI- alignment of proteins at the various CNR. structural levels, with the aim of finding common functional characteristics. Knowledge discovery and representation methods will be then exploited as knowledge-based adaptive systems for decision support in medicine and surgery (eg for studying autoimmunity mechanisms and for compatibility testing in organ transplant surgery).

Thanks to recent advances in microarray technology, we are now able to monitor the activity of a whole genome under

60 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

Numerical Mathematics Consortium

Scilab-publisher INRIA, National Instruments, Mathsoft and Maplesoft created the Numerical Mathematics Consortium and provide the foundations for a new generation of scientific software.

INRIA, National Instruments, Mathsoft and Maplesoft cre- INRIA brings a proven know-how in this field, thanks to ated the ‘Numerical Mathematics Consortium’ (NMC). These Scilab, a free open source numerical computation software mathematical software publishers as well as individuals from produced by the Scilab Consortium hosted by the Institute. industry and academia are collaborating to define a consistent and manageable foundation for numerical mathematics.

Announced initially in August in the USA, the NMC has been announced in Europe on November 17 by INRIA which is in charge of its development in Europe.

The initial objective of the Numerical Mathematics Consortium is to create an open mathematical semantics standard for numerical algorithm development to enable portability and reuse among tools, platforms and disciplines.

The Numerical Mathematics Consortium’s objective is to create a specification that stipulates core The Fast Fourier Transform (FFT)— a discrete Fourier transform algorithm — mathematical function definitions perfectly illustrates the need to standardize the semantic of functions. This and semantics applicable to numeric algorithm, applied to the same domain returns a different result according to whether algorithms. These algorithms can one uses Mathcad or Matlab/Scilab software. Both results are correct. Explanation: then be implemented in a wide vari- semantics differs! ety of application areas such as industrial control, embedded design and scientific research, as well as be easily shared among The software which is downloaded, each month, by more than researchers and developers in industry and academia. The 15,000 new users in the world, meets a growing success. group expects to complete a first release of the standard by next fall. The founding members of the Numerical Mathematics Consortium invite companies, research centres, universities Numerical calculation, simulation and control as well as and individuals who wish to take an active role in standardiz- embedded design are major disciplines for industry and ing numerical mathematics to join the Consortium. research. In these fields, incompatibility, unfortunately, pre- vails in a chronic way because, nowadays, semantics of math- Links: ematical functions can vary considerably from software to http://www.nmconsortium.org other. Each tool proposes its own set of functions, and it is http://www.scilab.org often necessary to rewrite the precious algorithms at the time of new projects or when new technologies are implemented. Please contact A standard set of mathematical functions based on common Didier Halgand, NMC marketing contact for Europe: semantics will allow both portable solutions and ready-to-use E-mail: [email protected] libraries and tools under numerous environments. The technological, scientific and financial stakes are thus more than sig- Claude Gomez, NMC scientific contact for Europe nificant. E-mail: [email protected]

ERCIM News No. 64, January 2006 61 R&D AND TECHNOLOGY TRANSFER

CORRECT – Developing Fault-Tolerant Distributed Systems

by Alfredo Capozucca, Barbara Gallina, Nicolas Guelfi, Patrizio Pelliccione and Alexander Romanovsky

The CORRECT project — rigorous stepwise development of complex fault-tolerant distributed systems: from architectural description to Java implementation — is an international project that aims to provide methodological support for developing fault-tolerant distributed systems in a cost-effective way. The proposed architecture-based methodology covers all phases of system development, from requirement specification to implementation. It is built on a Model-Driven Architecture methodology that automatically generates Java code from a UML- based software architecture description by means of semi-formal refinements.

Software and hardware systems have all the way to system implementation. exceptional behaviour unspecified. become widely used in many sectors, As graphically summarized in the figure, The dependability of such an SA spec- including manufacturing, aerospace, it consists of four steps: ification may be verified with SA- transportation, communication, energy related analysis tools. and healthcare. Failures due to software • Step 1: SA specification: initially, an or hardware malfunctions or malicious SA specification (described in some • Step 2: Annotated SA: software com- intentions can have economic conse- architectural language) is produced to ponents receive service invocations quences, but can also endanger human model architectural concepts such as and return responses when the services life. There is clearly a growing need for components, connectors, ports and are accomplished. The possible these systems to meet the highest configurations. This specification responses of a component when imple- dependability requirements. takes into account normal (ie mented and operating are ‘normal’ and expected) behaviour, while leaving ‘exceptional’. While normal responses Software Architectures (SA) involved report those situations where compo- in Model-Driven Engineering (MDE) nents provide normal services, techniques help to develop a system exceptional responses correspond to blueprint that can be validated and can errors detected in a component. guide all phases of system develop- Therefore, it is natural to design not ment. There are still many challenging only normal, but also exceptional issues in this area, in particular the behaviour. Similarly to normal integration of a new methodology into behaviour, exceptional behaviour the industrial software-development can be elicited from requirements life cycle, and in relating the results and thus modelled. The goal of this obtained by SA-based analysis to step is to extend the architectural requirements and coding. model in order to identify those components or connectors that raise and This situation has motivated the inves- handle exceptional behaviour. We tigation of a new architecture-centric annotate the architectural model with method for developing families of minimal information, only identify- complex fault-tolerant distributed sys- ing which component will handle tems. This effort is conducted by the each kind of exception. In contrast, Software Engineering Competence the exception-handling policies are Center at the University of identified later in the process by Luxembourg (SE2C) in collaboration employing notations and mecha- with the University of Newcastle upon nisms specific to fault tolerance. Tyne (UK), in the context of the COR- There are a number of reasons for RECT project. CORRECT is a three- this: the information available at the year (2004-2006) project, entirely architectural level could be insuffi- funded by the Luxembourg Ministry cient to identify how to handle the of Higher Education and Research. exception, and, since the handling of This architecture-based methodology exceptions depends strongly on the covers all phases of system develop- implementation language selected, ment, from requirements specification CORRECT methodology. we do not want to embed implemen-

62 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

tation-dependent information into the for handling exceptions. Since these Since it is usually impossible to generate architectural model. models are typically more informative a complete implementation, what we than the architectural ones, it is gener- may reasonably expect is to generate an • Step 3: Model-to-model transforma- ally impossible to derive a complete implementation schema with a set of tions: the essence of model-driven low-level fault-tolerance model. Some classes and their methods and exceptions development is the idea that a system extra information must be added and declaration. The role of a programmer is is incrementally obtained by instantiat- some refinements are needed. If the then to write the body of the methods and ing and refining a specification of sys- relevant architectural language sup- the code of the exceptional behaviour, tem structure and behaviour. The ports a behavioural description, this while the schema automatically manages OMG initiative Model-Driven information can be exploited in order exception propagation and orchestration. Architecture (MDA) is an implemen- to obtain a more detailed design tation of a more general trend of model. Language- or design-specific Link: model-driven development, with the exception-handling policies are http://se2c.uni.lu/tiki/tiki-index.php?pt= deployment effort focused around selected at this level, and are modelled Research%20Groups$ADS:%20Architecture$ industrial standards such as MOF, using design-level tools. CORRECT&page=CorrectOverview UML, CWM, QVT etc. • Step 4: Code generation via transfor- Please contact: Following the MDA principle, our mations: the next step involves the Nicolas Guelfi, University of Luxembourg methodology incorporates proper automatic generation of the applica- E-mail: [email protected] (automatic or semi-automatic) model- tion skeleton code from the fault-toler- http://www.ist.lu/users/nicolas.guelfi to-model transformation techniques to ance design model, via transformation generate more detailed design models rules.

The WearIT@work Project: Empowering the Mobile Worker with Wearable Computing by Elisa Basolu, Massimo Busuoli and Mike Lawo

The European Commission Integrated Project wearIT@work was set up to investigate the feasibility of ‘wearable computing’ as a technology to support the mobile worker. With a total budget of 23.7 million Euro, wearIT@work is the largest project in the world in this area. Based on four different pilot scenarios (emergency, variant production, maintenance and the clinical pathway) the project aims at a user-driven research agenda.

WearIT@work will prove the applicabil- the interface. This will capture their system and the environment. In addition, ity of computerized clothing, the so- attention. Thus, users can either interact there is direct interaction between the called ‘wearables’, in several industrial with the system or with the environment, system and the environment, and the sys- environments. The project is user-cen- but not with both at the same time. tem can mediate the interaction between tric, paying considerable attention to the the user and the environment. needs of the end users. A prime goal is to On the contrary, wearable systems allow investigate user acceptance of wearables. users to simultaneously interact with the However, four main issues must be addressed in order to implement wear- There are different approaches to wearable interaction concepts. able computing depending on the research direction and the application First, the system must be able to interact domain. In the wearIT@work project, a with the environment through an array of major focus is on the interactions sensors distributed in different parts of between user, system and environment. the outfit. In particular it must be able to develop a certain degree of awareness of In conventional mobile systems, the the user’s activity, physiological and interaction is based on a modified ver- emotional state, and the situation around sion of a desktop human computer inter- Interaction between the user, the system her/him. This is often referred to as con- face (HCI) and follows the pattern and the environment in a conventional text awareness. shown on the left side of the figure. To mobile system (left) and a wearable operate the system, users must focus on system (right).

ERCIM News No. 64, January 2006 63 R&D AND TECHNOLOGY TRANSFER

Second, the user interface needs to be Lastly, the system must be seamlessly tion. During the project lifetime, a call operated with minimal cognitive effort integrated in the outfit so that it neither will be issued for take-up projects to and little or no involvement of the hands. interferes with the users’ physical activ- adopt the solutions developed. More In general, a low cognitive load is ity nor affects their appearance in any information plus details on how to achieved through an appropriate use of unpleasant way. This means that, unlike receive the project newsletter with the the context information. Thus, for exam- many conventional mobile devices, it latest news on developments can be ple, instead of having the user select a can be worn nearly anywhere. found at the link below. function from a complex hierarchy of menus, the system should derive the two Nearly three years of project research The wearIT@work collaboration involves most preferable options from the context activity remain, but the fundamental 36 R&D groups (mainly from industry), information and present the user with a steps towards a user-centred design including ENEA, EADS, Skoda, HP, simple binary choice. In terms of the approach, a hardware framework and Microsoft, NTT DoCoMo, SAP, Siemens, actual input modality, simple natural software platform have already been Thales and Zeiss. The project is coordi- methods such as a nod of the head, a sim- achieved. With the creation of the Open nated by the Mobile Technology Research ple gesture, or spoken commands are Wearable Computing Group and the Center (TZI) of the University of Bremen. preferred. annual International Forum on Applied Wearable Computing, a community Link: Third, the system should be able to per- building process in industry and science http://www.wearitatwork.com. form a wide range of tasks using context has also been initiated. Although minia- information, without any user interaction turized and low-power computing Please contact: at all. This includes system self-configu- devices (as well as ubiquitous wireless Elisa Basolu, Massimo Busuoli, ration tasks as well as automatic communication) are still at an emerging ENEA - Italian National Agency for New retrieval, delivery, and recording of stage, it is our intention to exploit the Technologies, Energy and the Environment information that might be relevant to the project in order to encourage the wide E-mail: [email protected], user in a specific situation. A trivial adoption of wearable computing tech- [email protected] example of a context-dependent recon- nology. We aim to provide a wide spec- figuration could be a mobile phone that trum of innovative solutions in order to Mike Lawo, University of Bremen, Germany automatically switches off the ringer enable wearable computing to be used at E-mail: [email protected] during a meeting. any time, in any place and in any situa-

PULSERS Delivers on Phase 1 – Europe to Adopt a Ruling for Ultra-Wideband

by Walter Hirt

The Electronic Communications Committee (ECC) in the European Conference of Post and Telecommunications Administrations (CEPT) recently completed its public consultation on ‘Harmonized conditions for devices using UWB technology in bands below 10.6 GHz’. This action represents an important step towards the introduction of a legal framework in Europe for this emerging – sometimes Pervasive Ultra-wideband considered ‘disruptive’ – ultra-wideband radio technology (UWB-RT). Members Low Spectral Energy of the IST-FP6 Integrated Project PULSERS, which is entirely focused on the Radio Systems advancement of UWB-RT, played an active role in the European regulatory process that led to this milestone result.

The IST-FP6 Integrated Project less communication and ranging and with location tracking (LDR-LT), cover- PULSERS is a consortium of thirty localization applications, in both tradi- ing distances up to tens of metres (see industrial and academic organizations, tional and novel use scenarios. When Figure 1). which was formed to promote UWB-RT: trading data rate versus range, the tech- ‘Optimally Connected Anywhere and nology accommodates two complemen- PULSERS conceived HDR/VHDR sys- Anytime’. In Phase 1 (2004–2005) of the tary classes of systems: (i) systems offer- tems to support a number of applications project, PULSERS introduced and ing high data rates (HDR) or very high in the home and office environments, developed new concepts for future short- data rates (VHDR) over links of up to a such as wireless video connections and range wireless systems. The underlying few metres, and (ii) systems supporting wireless high-speed connections for UWB-RT is capable of supporting wire- low data rates (LDR) alone or combined computing equipment. The newly intro-

64 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

duced class of LDR-LT devices combines low-rate data-transfer capabilities with precise ranging and location tracking, particularly for indoor environments; these features are increasingly being requested in wireless sensing and control applications. During the first two years of the project, PULSERS established significant results in the following technical areas: • User Application Scenarios – user scenarios and business applications applying the unique technical properties of UWB-RT were defined and their market potential was assessed. • System Concepts – single and multiple antenna systems (SAS, MAS) were designed or investigated. SAS designs Figure 1: Envisaged application scenarios for short-range wireless systems based on partly benefited from the results of UWB-RT (see also D. Porcino and W. Hirt, ‘Ultra-Wideband Radio Technology: Potential UWB-related FP5 projects, whereas and Challenges Ahead,’ IEEE Commun. Mag., July 2003, pp. 66-74; the WBAN fundamental theoretical investigations illustration is courtesy of the WWRF). were necessary for MAS. • PHY Concepts – novel physical layer (PHY) concepts for HDR/VHDR and with existing and future (viz. 4G) radio- The prospects for arriving at a commer- LDR-LT devices were developed and communication services. cially viable UWB-spectrum regulation issues related to their interoperability in Europe are therefore intact, albeit not explored and assessed. Members of PULSERS have addressed to the same extent as in the United States • MAC Concepts – medium access con- these legitimate concerns through tech- of America, where the legal use and mar- trol (MAC) functions were modelled nical studies, and the results have been keting of UWB-RT was authorized in and designed for HDR and LDR-LT made available to European and other early 2002. A technically and commer- devices, with an emphasis on cross- international bodies engaged in UWB- cially sound regulatory framework for layer issues between MAC and PHY. related harmonization processes. This UWB-RT needs to be established in • Enabling Technologies – high-speed, required close cooperation with key Europe: one that also accommodates the mixed-signal semiconductor processes external industrial organizations promot- protection needs of affected spectrum were exploited to verify the designs of ing the deployment of UWB-RT (eg stakeholders. This is a key prerequisite critical analogue/digital building WiMedia Alliance and UWB Forum). In for the successful introduction of harmo- blocks; UWB antenna designs were addition, members of PULSERS main- nized technical standards and essential advanced. tained a constructive liaison with the for establishing a viable European European Commission’s DG INFSO B4 ecosystem based on UWB-RT. A Regulation and Standards (Radio Spectrum Policy). DG INFSO B4 favourable status quo of UWB-RT in A specific work package in PULSERS holds the authoritative stakes for the Europe and beyond would provide sig- addressed the complex issues related to final decision on the regulatory condi- nificant benefits to consumers and busi- the spectrum regulation and technical tions that will govern the deployment of nesses alike and would help to sustain standardization of UWB-RT. UWB-RT UWB-RT in the European Union. It was the success of PULSERS in Phase 2. is often referred to as a ‘disruptive’ radio significant that the Electronic technology because the premise of Communications Committee in the Links: UWB-RT is founded on reusing fre- CEPT (European Conference of Post and http://www.pulsers.net quency bands already assigned to incum- Tele-communications Administrations) http://www.cordis.lu/ist/ct/proclu/p/projects.htm bent radiocommunication services. This recently completed its public consulta- is done by spreading signals of very low tion on the ‘Harmonized conditions for Please contact: power (ie less than one thousandth of the devices using UWB technology in bands Walter Hirt power emitted by a cellular phone) over below 10.6 GHz’. While the final out- IBM Research GmbH, Switzerland bandwidths of up to several Gigahertz. come of this action is still pending, this Tel: +41 44 724 8477 Understandably therefore, spectrum- action represents an important step E-mail: hir@zurich.ibm.com granting administrations, licence-hold- towards the introduction of a legal ing operators and other stakeholders of framework in Europe for UWB-RT, spectrum resources have expressed their which is now scheduled for mid 2006. concerns that UWB-RT will interfere

ERCIM News No. 64, January 2006 65 R&D AND TECHNOLOGY TRANSFER

AVISPA: Automated Validation of Internet Security Protocols and Applications

by Alessandro Armando, David Basin, Jorge Cuellar, Michael Rusinowitch and Luca Viganò

AVISPA is a push-button tool for the Automated Validation of Internet Security Protocols and Applications. It provides a modular and expressive formal language for specifying protocols and their security properties, and integrates different back-ends that implement a variety of state-of-the-art automatic analysis techniques. Experimental results, carried out on a large library of Internet security protocols, indicate that the AVISPA tool is the state of the art for automatic security protocols. No other tool combines the same scope and robustness with such performance and scalability.

With the spread of the Internet and net- mistakes in protocol design. Typically small and medium-scale protocols have work-based services and the develop- these attacks go unnoticed, as it is diffi- been developed, moving up to large-scale ment of new technological possibilities, cult for humans, despite careful protocol Internet security protocols remains a the number and scale of new security pro- inspection, to determine all the complex challenge. The AVISPA tool is a push- tocols under development is outpacing ways in which protocol sessions can be button tool for the Automated Validation the human ability to rigorously analyse interleaved, with the possible interfer- of Internet Security-sensitive Protocols and validate them. This is an increasingly ence of a malicious intruder. and Applications, which rises to this serious problem for standardization orga- challenge in a systematic way. First, it nizations like the Internet Engineering Tools that support a rigorous analysis of provides a modular and expressive for- Task Force (IETF), the International security protocols are thus of great mal language for specifying security pro- Telecommunication Union (ITU) and the importance in accelerating and improv- tocols and properties. Second, it inte- World Wide Web Consortium (W3C). It ing the development of the next genera- grates different back-ends that implement also affects companies whose products tion of security protocols. Ideally, these a variety of automatic analysis tech- and services depend on the rapid stan- tools should be completely automated, niques ranging from protocol falsifica- dardization and correct functioning of robust, expressive and easily usable, so tion (by finding an attack on the input these protocols, and users whose rights that they can be integrated into protocol protocol) to abstraction-based verifica- and freedoms (eg the right to privacy of development and standardization pro- tion methods for both finite and infinite personal data) depend on a secure infra- cesses. numbers of sessions. To the best of our structure. knowledge, no other tool exhibits the Although in the last decade many new same scope and robustness while enjoy- Designing secure protocols is a hard techniques that can automatically analyse ing the same performance and scalability. problem. In open networks such as the Internet, protocols As shown in the figure, should work even under AVISPA is equipped with a worst-case assumptions, eg Web-based graphical user that messages may be seen or interface that supports the edit- tampered with by an intruder ing of protocol specifications (also called the attacker or and allows the user to select spy). Severe attacks can be and configure the back-ends conducted without breaking integrated into the tool. If an cryptography, by exploiting attack on a protocol is found, weaknesses in the protocols the tool displays it as a mes- themselves. Examples of this sage-sequence chart. The are ‘masquerading attacks’, in interface features specialized which an attacker imperson- menus for both novice and ates an honest agent, or ‘replay expert users. A protocol attacks’, in which messages designer interacts with the tool from one protocol session (ie by specifying a security prob- execution of the protocol) are lem (ie a protocol paired with a used in another session. The security property that the pro- possibility of these attacks AVISPA Web-based tocol is expected to achieve) in sometimes stems from subtle graphical user interface. the High-Level Protocol

66 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

Specification Language (HLPSL). The large subset of these protocols in ware distribution. For more details, HLPSL is an expressive, modular, role- HLPSL. The result of this specification please refer to the AVISPA Web site. based, formal language that is used to effort is the AVISPA Library (publicly specify control-flow patterns, data-struc- available on the AVISPA Web site), AVISPA has been developed in the con- tures, alternative intruder models and which at present comprises 215 security text of the FET Open Project IST-2001- complex security properties, as well as problems derived from 48 protocols. 39252 ‘AVISPA: Automated Validation different cryptographic primitives and Most of the problems in the library can of Internet Security Protocols and their algebraic properties. These fea- be solved by the AVISPA tool in a few Applications’, in collaboration with the tures make HLPSL well suited for speci- seconds. Moreover, AVISPA detected a University of Genova, INRIA Lorraine, fying modern, industrial-scale protocols. number of previously unknown attacks ETH Zurich and Siemens Munich. on some of the protocols analysed, eg on In order to demonstrate the effectiveness some protocols of the ISO-PK family, Link: of AVISPA, we selected a substantial on the IKEv2-DS protocol, and on the http://www.avispa-project.org set of security problems associated with H.530 protocol. protocols that have recently been, or are Please contact: currently being standardized by organi- The AVISPA tool can be freely accessed Alessandro Armando, zations like the Internet Engineering either through its Web-based interface or Università di Genova, Italy Task Force IETF. We then formalized a by downloading and installing the soft- Tel: +39 010353 2216 E-mail: [email protected]

Prague Texture Segmentation Data Generator and Benchmark by Stanislav Mikeš and Michal Haindl

The Prague texture segmentation data-generator and benchmark is a Web-based service developed as a part of the MUSCLE Network of Excellence. It is designed to mutually compare and rank different texture segmenters and to support new segmentation and classification methods development. It can be easily used for other applications such as feature selection, image compression, and query by pictorial example.

Unsupervised or supervised texture seg- facing similar problems during our tex- The benchmark allows users to: mentation is the prerequisite for success- ture segmentation research, and for this • obtain customized texture mosaics (U) ful content-based image retrieval, auto- reason we implemented a solution in the or benchmark sets and their corre- matic acquisition of virtual models, qual- form of a Web-based data generator and sponding ground truth ity control, security, medical applica- benchmark software. • evaluate working segmentation results tions and many others. Although many and compare them with state-of-the- methods have already been published, The goal of the benchmark is to produce art algorithms this problem is still far from being a score for an algorithm’s performance. • update the benchmark database (U) solved. This is partly due to the lack of This is done so that different algorithms with an algorithm (reference, abstract, reliable performance comparisons can be compared, and so that progress benchmark results) and use it for the between the different techniques. Rather toward human-level segmentation per- benchmarking of subsequent algo- than advancing the most promising formance can be tracked and measured rithms image segmentation approaches, over time. The benchmark operates • check single mosaic evaluation details researchers often publish algorithms that either in full mode for registered users (criteria values and resulting thematic are distinguished only by being suffi- (unrestricted mode - U) or in a restricted maps) ciently different from previously pub- mode. The major difference is that the • rank segmentation algorithms accord- lished algorithms. The optimal alterna- restricted operational mode does not ing to the most common benchmark tive is to check several variants of a store a visitor’s data (results, algorithm criteria and receive the resulting method being developed and to carefully details etc) in its online database, and LaTeX-coded criteria tables (U). compare results with the state of the art. does not allow custom mosaics creation. Unfortunately, this is impractical, since To be able to use the complete and unre- Benchmark datasets are computer-gen- most methods are too complicated and stricted benchmark functionality, regis- erated random mosaics filled with ran- insufficiently described to be imple- tration is required. domly selected textures. Both generated mented with acceptable effort. We were texture mosaics and the benchmarks are

ERCIM News No. 64, January 2006 67 R&D AND TECHNOLOGY TRANSFER

thematic classes. The bench- We have implemented the most fre- mark uses cut-outs from the quented nineteen evaluation criteria cat- original textures (1/6 approxi- egorized into three groups: region-based mately) either in the original (5), pixel-wise (12) and consistency resolution or a sub-sampled measures (2). The performance criteria version. The remaining parts mutually compare ground truth image of the textures are used for regions with the corresponding machine- separate test or training sets in segmented regions. The region-based the benchmark-supervised criteria are correct, over-segmentation, mode. The benchmarks use under-segmentation, missed and noise. 114 colour/greyscale textures Our pixel-wise criteria group contains from ten classes. The BTF the most frequented classification crite- measurements are provided ria such as omission and commission Figure 1: Benchmark colour textures. courtesy of Prof. Reinhard errors, class accuracy, recall, precision Klein from Bonn University. etc. Finally the last criteria set incorporates the global and local consistency Colour, greyscale or BTF errors. The evaluation table is re-ordered benchmarks are generated according to the chosen criterion. For upon request in three quantities each compared algorithm there is a con- (normal=20, large=80, cise description available, eg author, huge=180 test mosaics). For algorithm details, BIB entry and WWW Figure 2: Example of a texture mosaic, ground truth each texture mosaic the corre- external page. and segmentation result. sponding ground truth and mask images are also included. Links: The test mosaic layouts and http://mosaic.utia.cas.cz composed from the following texture each cell texture membership are ran- http://www.utia.cas.cz/RO types: (i) monospectral textures (derived domly generated, but identical initializa- BTF measurements (courtesy of Prof. from the corresponding multispectral tion of the corresponding random genera- Reinhard Klein): textures), (ii) multispectral textures, and tors is used, so that the requested bench- http://btf.cs.uni-bonn.de/index.html (iii) BTF (bi-directional texture func- mark sets (for the same size and type) are tion) textures. identical for each visitor. Please contact: Stanislav Mikeš, Institute of Information The benchmark uses colour textures The submitted benchmark results are Theory and Automation, Academy of from our Prague colour texture database, evaluated and stored (U) in the server Sciences / CRCIM, Czech Republic which contains over 1000 high-resolu- database and used for the algorithm Tel: +420-266052350 tion colour textures categorized into ten ranking according to a chosen criterion. E-mail: xaos@utia.cas.cz

Visiting Hybrid Museums: A Colony of Ants in your Pocket

by Javier Jaén and José A. Mocholí

Tangible and intangible digital objects in areas of cultural heritage are becoming increasingly interrelated. Intangible multimedia objects provide contextualization and additional information that aids in the understanding of artistic creative processes. These types of interrelated information must therefore be made available for museum visitors, so that they may have a more enriching experience. This note reports on our experience in the use of evolutionary algorithms based on ant colonies for the efficient provision of dynamic time-constrained visits in Hybrid Museums.

One of the most enriching and exciting result, in the cultural domain, there is a dangers such as forgetting the past and experiences that differentiates human mandate to save not only the artistic destroying memory. beings from other animals is our ability expressions from earlier times, but also to use expressions of artistic or cultural the commentaries, reflections and In this respect, digital culture must face a work to stimulate our senses and to knowledge relationships relating to number of challenges in the coming experience a range of emotions. As a these expressions. In this way, we avoid decades. First, it must provide adequate

68 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

multimedia expressions for both tangible the trail pheromone. This chemical is and intangible heritage. Second, these used to mark paths on the ground from new multimedia and multimodal forms food sources to the nest. Several experi- of culture must be properly conserved ments have shown that this communica- and preserved. Third, both tangible and tion mechanism is very effective in find- intangible cultural expressions must be ing the shortest paths and as a result, has properly interrelated with respect to dif- inspired several stochastic models that ferent awareness contexts or dimensions. describe the dynamics of these colonies. Lastly, the resulting vast amount of distributed and interrelated cultural infor- Ant Colonies optimization algorithms mation must be properly delivered to the are effective heuristics for solving prob- end users. This should be done with sim- lems of medium size (hundreds of ple navigation or exploration mecha- nodes). However, the computational cost nisms that are intuitive and hide the when dealing with the thousands of inherent complexity of the different nodes of a museum is prohibitive, forms of interrelated heritage. because visitors are not willing to wait several minutes (even half an hour) to Figure 1: Ant Colony solution at El Prado The MoMo project, a collaboration obtain a reasonable solution. Instead, our Museum. between the Polytechnic University of strategy has been to solve large OP Valencia in Spain and the Microsoft instances with thousands of nodes by Research Labs (Cambridge), is a step partitioning the search space into sub- towards solving these challenges in the spaces and then solving the subprob- context of hybrid museums (HMs). HMs lems. This is done with the help of a Grid are known as infrastructures that enable computing infrastructure with no com- the exploration of traditional museums munication for synchronization purposes with the assistance of wireless Personal among the worker nodes of the infra- Digital Assistants (PDAs) that have mul- structure. Our mechanism can be seen as timedia capabilities and are able to adapt a master-slave approach like that pro- dynamically to visitors’ preferences and posed by Middendorf, but with no behaviour. However, as the number of pheromone matrix propagation because cultural elements and amount of interre- the slaves work on independent lated information grows, visitors need instances. some form of automatic assistance in deciding how to visit as many popular The results we have obtained prove that artworks (those belonging to the most instances of this problem of up to several relevant artists) as possible within their thousand elements can be solved within available time. a few seconds. Moreover, because we have implemented this distributed infra- This problem, which has traditionally structure with the .NET technology, not been known as the Orienteering Problem only desktop PCs but also the handheld Figure 2: MoMo’s Graphical User (OP), is a combinatorial problem that devices that are present in the HM can Interface. cannot be solved in polynomial time. host ant colonies.. Figure 1 shows an Moreover, in the case of museums where example of a computation obtained for nodes in the OP graph are artworks, the El Prado museum, where the red rooms sizes of typical OP instances are in the are the most attractive ones and the blue thousands of nodes. Therefore, solving ones are the least. It can be observed that Links: such instances with exact algorithms is the collective effort of ants obtains a path http://momo02.dsic.upv.es not a feasible approach. Instead, mecha- that visits most of the popular rooms http://iridia.ulb.ac.be/~mdorigo/ACO/ACO.html nisms based on heuristics are more suit- within the specified available time. http://research.microsoft.com/aboutmsr/labs/ able. In particular, our research makes cambridge/default.aspx use of evolutionary algorithms inspired This evolutionary approach, together by the natural behaviour of ant colonies with some additional intelligent predic- Please contact: to provide efficient and high-quality tion components, contributes to more Javier Jaén, solutions to this problem. Ant colonies exciting and enriching museum experi- Universidad Politécnica de Valencia / SpaRCIM are insect societies that accomplish com- ences and could be the foundation for E-mail: [email protected] plex tasks by presenting highly struc- museums that provide more effective http:// issi.dsic.upv.es/Members/plone/ tured organizations and communication guidance to visitors in the near future. Members/profes/fjaen mechanisms. Ant behaviour is based on the use of pheromones, and in particular

ERCIM News No. 64, January 2006 69 R&D AND TECHNOLOGY TRANSFER

Edutainment and Game Theory Realization through an Open-Source UNESCO Virtual Perspolis Project

by Sepideh Chakaveh, Stefan Werning and Olaf Geuer

The ‘Virtual Persepolis’ project applies current online game theories to a historical event. The original idea was developed in a Master thesis at at the Fraunhofer Institute for Media Communication and the ERCIM E-Learning Working Group.

This year the Nobel Prize in Economic Economics is not the only subject associ- course. However, a lack of resources or Sciences was awarded to two game theo- ated with game theory: psychology is other factors may limit the situations in rists. Robert J. Aumann, an Israeli- another field within which, through the which this is possible. In contrast, a com- American, and Thomas C. Schelling, a theory of social situations, game theory puter simulation of such a competition US citizen, defined chess-like strategies is interwoven. Although here the rele- would enable more rapid prototyping and in politics and business that can be vance is more to parlour games such as further refinement, and could expand the applied to arms races, price wars and poker or bridge, most research focuses total number of students who can prop- actual warfare. Game theory is a distinct on how groups of people interact. erly share in the experience. Games may and interdisciplinary approach to the also enable teachers to observe their stu- study of human behaviour. The disci- The strength of game theory lies in the dents’ problem-solving strategies in plines most involved in game theory are fact that no parameter is fixed. Only the action and to assess their performance in mathematics, economics and the other boundaries are predefined, since deci- realistic situations. Teachers may also social and behavioural sciences. Game sions must be made in accordance with demonstrate a particularly difficult prob- theory (like computational theory and so the status quo. This idea is not new in lem during a lecture and discuss possible many other contributions) was founded computer science, as it defines a solutions. After all, demonstrating the by the great mathematician John von quintessential element of decision-mak- principles of Newtonian physics using Neumann. The first important book on ing in almost all autonomous systems. A gears, pulleys and levers may be more the subject was The Theory of Games simple analogy is a robot moving on a compelling than chalk on the blackboard. and Economic Behaviour, which von course: when it encounters an obstacle, it It is not just that games can help one do Neumann wrote in collaboration with the must decide which route to take in order better on the test; games could become great mathematical economist, Oskar to arrive at its final goal. In some cases the test. Morgenstern. the robot may also decide whether to move over the obstacle, if the dimen- Virtual Persepolis Project Principals of Game Theory sions or materials permit. The original idea for this project was The application of game theory in eco- first developed in a Masters thesis in nomics relies on the assumption that Edutainment and Game Theory 2003, at the Fraunhofer Institute IMK individuals are absolutely rational in In the recent years, educational research and the ERCIM E-Learning Working their economic choices. In particular it is has shown that peer-to-peer teaching Group. In this research, MULE (Multi- assumed that each person maximizes her reinforces mastery. Moreover, educators User Learning Environment) systems or his rewards – that is, profits, incomes have recognized the value of practical and actual online game theories are or subjective benefits – in the circum- experience and competition. For applied to historical events. The Virtual stances that she or he faces. This hypoth- instance, students can design and build Persepolis project was implemented esis serves a double purpose in the study robots which then compete against one through an appropriate Internet 3D of the allocation of resources. other in navigating through an obstacle Game Engine. The visualization centred on a reconstruction of part of the Apadana Palace.

In this scenario, a tutor can use rich multimedia tools (eg pictures, 2D and 3D animations, maps, music, audio and video) to illustrate architectural, historical and cultural facts. The players of the game (in this case represented by Persian Guards) are simulated by avatars, each of which bears a national and cultural Persepolis Project screenshots. identity described by the predefined

70 ERCIM News No. 64, January 2006 R&D AND TECHNOLOGY TRANSFER

parameters of game theory. Persian sol- parameters and boundaries are assumed scenarios may be simulated by anyone diers with appropriate decision-making to be accurate and realistic. This implies who has access to the Internet. This pro- capabilities were realized through that the decisions made by virtual charac- vides a unique and powerful tool for Artificial Intelligence (AI), which are ters (or players) based on AI can only be learning, planning and experiencing variables in this programme. This could correctly determined when the defined events which until very recently were for instance depend on which side of the parameters are realistic. limited to the very privileged. Apadana Palace one would face them and which route they take in the palace. Tying advances in Web-based edutain- Link: ment systems to the sound and challeng- http://mats.imk.fraunhofer.de/e-learning/ The principle of collaborative learning ing concepts of game theory allows through role playing and interactive vir- interesting areas of research to be Please contact: tual systems allows the user to experience explored that are not bounded by com- Sepideh Chakaveh, ERCIM E-Learning and an event as it unfolds, and thus learn to puter science alone. With the application Edutainment Working Group coordinator make appropriate decisions on the fly. of low-end virtual reality and artificial Institute for Media Communication, This is of course only true when the input intelligence, a number of long-awaited Fraunhofer ICT Group

CONFIOUS: Conference Management System with Intelligence, Power and Style by Manos Papagelis and Dimitris Plexousakis

Confious is a state-of-the-art conference management system that combines Notification modern design, sophisticated algorithms and a powerful engine to efficiently support the submission and reviewing process of an academic conference or Submission Revision workshop and to help Program Committee Chairs and members to effortlessly carry out complicated tasks.

Most scientific communities have ences in parallel, transparent between recently established policies and mecha- each other. Moreover, there is no need Review nisms to put into practice electronic con- for installations on the conference orga- Bidding ference management, mainly by exploit- nizer side. Quite the opposite, a confer- ing the Internet as the communication ence is hosted on the server side, and it and cooperation infrastructure. Their can accept submissions immediately foremost objective is to minimize the after customization. organizational effort and reduce commu- Submission and reviewing process of an nication costs, while maintaining high 100% Online, Role-Based Collaboration: academic conference or workshop. quality reviewing and ensuring the fair- Confious distinguishes between four ness of the evaluation. With the intention types of users: Program Committee automatically or manually. The most of supporting scientific committees in Chair, Meta-Reviewer, Reviewer and advantageous process includes an auto- this process (see figure), we have Contact Person. matic assignment by the system fol- designed and implemented Confious, a lowed by manual adjustment of assign- state-of-the-art conference management Intelligent Management of Conflicts of ments by the chair. The automatic system. Confious provides mechanisms Interest: A conflict of interest is defined assignment algorithm takes into consid- for the efficient management of scien- when a reviewer does not feel comfort- eration the following constraints: tific data, intelligent identification and able reviewing a paper because of some • matches between paper topics and analysis of constraints during the kind of relationship with one or more of reviewer interests reviewing process, enhanced monitoring the paper’s authors. Confious provides • bids of reviewers to specific papers and reporting of critical actions and bet- an easy way to effectively identify and • conflicts of interest between reviewers ter communication among users. handle all conflicts of interest and and papers thereby avert certain assignments. • workload balance to ensure that papers Main Features are normally distributed to reviewers. Instant Conference Setup: Confious is Automatic Assignment of Papers to provided as a Web service, which means Reviewers: Program Committee Chair Dynamic Construction of the Review that it is able to support multiple confer- has the option of assigning papers either Form: The online review form is com-

ERCIM News No. 64, January 2006 71 R&D AND TECHNOLOGY TRANSFER

pleted by reviewers during the reviewing classes and then orders the papers in Task-Oriented Design: Confious is based process. Instead of using static review each class separately. on a goal-oriented design that facilitates forms that have the same format, the management and administration of a Confious provides the Chair with the Workflow Management through Phases: conference, enhances productivity and option of constructing or customizing a The reviewing and submission process is advances creativity. Special attention has review form. This is one of the most divided into several phases that enable been paid to effortlessly bringing tasks to advantageous features of Confious since more efficient management and monitor- an end, providing quick responses to user it enables high-quality reviewing and yet ing of the tasks that are being performed requests and permitting access even from is not supported by most other confer- at any given moment. low-speed connections. ence management systems. Enhanced Monitoring, Reporting and Commentary: We are confident that the Hierarchical Reviewing Support: The Communication Service: Confious pro- rational and algorithmic ground on which distinction between reviewers and meta- vides a monitoring and reporting service we have designed and developed reviewers enables a hierarchical review- that spans the most essential tasks of the Confious will catch the attention of con- ing process. This process ensures high- system. Furthermore, straightforward ference organizers, and will exert a pull quality reviewing by providing a better customization of e-mails enables better on scientific committees to consider it for means of control over the submitted communication between the Chairs, the their future conferences. The interested reviews. reviewers and the authors. reader is encouraged to visit Confious online, in order to obtain a better under- Decision-Making Based on Paper Transaction-Based Operations: Data standing of the system’s features through Classification: One of the most chal- inconsistencies are very common to Web a demonstration conference or through an lenging and time-consuming tasks that information systems. They can arise due up-to-the-minute trial conference. the Chair is in charge of is to decide on to network instabilities, power cut-offs the sets of papers that are going to be or other reasons and may result in seri- Link: accepted or rejected. Actually, it is hard ous conflicts during the reviewing pro- http://confious.ics.forth.gr to reduce the results of several reviews cess. Database operations in Confious into a single meaningful score. Confious are based on transactions to ensure that Please contact: classifies the papers into five meaningful data always remains consistent. Manos Papagelis, ICS-FORTH, Greece E-mail: [email protected]

Advertisement Call for Take up Actions AXMEDIS Project of the European Commission — http://www.axmedis.org

The Integrated Project AXMEDIS - Automating next AXMEDIS conference http://www.axmedis.org/axmedis2006 Production of Cross Media Content for Multi-channel Distribution - The candidate topic areas of this call include the followings application requires the participation of new contractors to carry out take-up and/or extension of the AXMEDIS framework and tools to support: actions as sub-projects within AXMEDIS project to promote the • one or more distribution channels in order to make evident validation and early application of AXMEDIS technologies via interoperability of content and tools with other AXMEDIS demonstration activity. distribution channels and tools (mobile devices, PC, STB, portable video player, portable music player, etc.) AXMEDIS is providing a framework which includes technologies, methods and tools to speed up and optimise content production, • massive and/or coordinated production, aggregation, protection, protection and distribution, for leisure, entertainment and digital of cross media content; collaboration among different actors of content valorisation and exploitation in general for multi-channel the production and distribution value chain; collaborations among distribution, supporting interoperability of content and DRM. cultural institutions, etc. AXMEDIS aims to meet the challenges of digital-content market • production and/or distribution authoring tools and/or players. demand by: (i) reducing costs for content production and management by applying composition, parallel processing, Take up projects should aim at developing real solutions (adoption optimisation techniques for content formatting and representation of the AXMEDIS framework and technology in real-life scenarios) by (format) and workflow control; (ii) reducing distribution and exploiting AXMEDIS technologies. They should start real aggregation costs in order to increase accessibility with a Peer-to- sustainable activities by taking advantage of the AXMEDIS Peer platform at Business-to-Business level, which can integrate framework services and derived tools. Maximum funding is about 1- content management systems and workflows; (iii) providing new 1.1 million Euro, for the whole 3-4 take up actions. methods and tools for innovative, flexible and interoperable Digital Rights Management (DRM), including the exploitation of MPEG-21 All the necessary information for submitting your proposal is and overcoming its limitations, and supporting different business available at the call webpage of the AXMEDIS project and transactions models. http://www.axmedis.org/callfortakeup/call.html For the technical details regarding AXMEDIS framework specification please visit AXMEDIS web site on which Tutorials, Contact: Specification, Use Cases, Test Cases, and reports about the Prof. Paolo Nesi, AXMEDIS project co-ordinator research activity performed and planned are available. See also E-mail: [email protected]; http://www.dsi.unifi.it/~nesi/

72 ERCIM News No. 64, January 2006 EVENTS

EU-INDIA’05 - Second Annual Conference 7th Workshop of the ICT for EU-India Cross Cultural of the ERCIM Dissemination Project Working Group by María Alpuente, Santiago Escobar, and Moreno Falaschi ‘Matrix Computations The Second Annual Conference of the ‘ICT for EU-India Cross Cultural and Statistics’ Dissemination Project’, EU-INDIA’05, was held in Valencia, Spain, 14-15 November 2005. The workshop was organized within the framework of the 3rd World conference In the framework of the EU-India on Computational Statistics and Data Economic Cross Cultural Programme, Analysis which was held in Limassol, the ‘ICT for EU-India Cross Cultural Cyprus, 28-31 Octobre2005. Chairman Dissemination Project’, coordinated by of the conference was Erricos Prof. Furio Honsell from University Kontoghiorghes from the university of Udine, aims at creating a network for Cyprus. developing collaborations and exchanges between European and Indian Two sessions were organized by the universities and R&D research centres in working group: the areas of integrating new communica- • Data assimilation and its application: tion technologies in education and dis- (organized by Z. Zlatev, NERI, semination of entrepreneurial research Denmark) from academia to business, with particu- Speakers: V. Mallet, G. Dimitriu, Z. lar focus on heritage and e-governance. Zlatev, F-X Le Dimet, S. Zein. • QR and other factorizations (orga- The conference was attended by 30 par- nized by B. Philippe, INRIA/IRISA, ticipants from universities and research From left: B.G. Sidharth (Birla Science France) institutes of the four partner universities Centre, Hyderabad), María Alpuente (U. Speakers: L. Grigori, F. Guyomarc’h, Udine, Genoa (both Italy), Valencia Politécnica de Valencia) and Furio D. di Serafino, J. Barlow. (Spain), and Hyderabad (India). The Honsell (University of Udine). conference was inaugurated by Links: Gumersindo Verdú, vice-chancellor for The title of the talks and the slides: international actions of the Technical repairing, Web-based declarative con- http://www.irisa.fr/sage/wg-statlin/ Unversity of Valencia (UPV). The three necting tools, digital platforms for artis- WORKSHOPS/LEMASSOL05/program.html invited speakers gave highly interesting tic collections, and hybrid museum infra- and stimulating presentations. Jaime structure. WG home page: Gómez, general director for scientific http://www.irisa.fr/sage/wg-statlin/ and technological parks of the Valencian The conference chairs wish to thank the government presented the ‘Valencian different institutions and corporations Please contact: Government Model for Research and that have supported the event, especially Erricos Kontoghiorghes, Science Parks’. Juan José Moreno the EU-India project ALA/95/23/ University of Cyprus Navarro, international relations head for 2003/077-054, Università degli Studi di E-mail: [email protected] IST of the spanish ministry of education Siena, Technical University of Valencia, and science, described the ICT research and the Spanish Ministry of Education Bernard Philippe, INRIA, France activities in Spain and their European and Science. E-mail: [email protected] and Indian connections. Finally, Salvador Coll, director for new initia- Link: tives of the UPV ‘Iinnovation Technical EU-INDIA’05: City’ (CPI) in Valencia, spoke about http://www.dsic.upv.es/workshops/euindia05/ research, innovation, challenges, and achievements at the CPI Science Park. Please contact: Maria Alpuente, Technical University of The technical program also included fif- Valencia / SpaRCIM, Spain teen regular talks organized in six ses- Tel: +34 963 879 354 sions: web categorization, natural lan- E-mail: [email protected] guage engineering, Web verification and

ERCIM News No. 64, January 2006 73 EVENTS

ERCIM News is the magazine of ERCIM. CALL FOR PARTICIPATION ERCIM-Sponsored Events Published quarterly, the newsletter reports on joint actions of the ERCIM partners, and aims to reflect CISTRANA Workshops Series ERCIM sponsors up to ten the contribution made by ERCIM to the European conferences, workshops and Community in Information Technology. Through on Coordination of IST Research summer schools per year. The short articles and news items, it provides a forum funding for all types of events is for the exchange of information between the and National Activities institutes and also with the wider scientific 2000 Euro. community. This issue has a circulation of 10,500 A series of workshops is being organised copies. The printed version of ERCIM News has in the scope of the CISTRANA project Conferences a production cost of 8 Euro per copy. It is available ERCIM invites sponsorship proposals from free of charge for certain groups. (http://www.cistrana.org), whose aim is to achieve coordination of national established conferences with an Advertising For current advertising rates and conditions, research programmes with each other international reputation, where substantive see http://www.ercim.org/publication/ERCIM_News/ and with European programmes in the overlap can be shown between the or contact [email protected] information technology sector, in order conference topic and ERCIM areas of Copyright Notice to improve the impact of all research activity. Typical cases would include annual All authors, as identified in each article, retain efforts in Europe and to reinforce conferences in computer science with copyright of their work. European competitiveness. The work- international programme committees, ERCIM News online edition http://www.ercim.org/publication/ERCIM_News/ shops are open to anyone with an interest substantial international participation, and ERCIM News is published by ERCIM EEIG, in the topics. This might include pro- proceedings published with an established BP 93, F-06902 Sophia-Antipolis Cedex gramme and project managers, persons international science publisher. Tel: +33 4 9238 5010, E-mail: [email protected] interested in multinational collabora- ISSN 0926-4981 tions, developers of research portals, and Workshops and Summer Schools Director: Jérôme Chailloux, ERCIM Manager users of taxonomies in information tech- ERCIM sponsors workshops or summer Central Editor: nology. Leading speakers have been schools (co-) organised by an ERCIM Peter Kunz, ERCIM office [email protected] invited, and there will be opportunities institute. The additional funding provided Local Editors: for open discussion to reach conclusions by ERCIM should be used to enhance the AARIT: n.a. that will have a real impact on the future workshop by, for example, increasing the CCLRC: Martin Prime of the European research landscape. number of external speakers supported. [email protected] CRCIM: Michal Haindl [email protected] The future workshops are as follows: Next Deadlines for Applications: CWI: Annette Kik • Best practice in multi-national pro- • Conferences: 15 April 2006 for [email protected] gramme collaboration (18 Jan conferences later than 15 December CNR: Carol Peters [email protected] 2006, Cologne, Germany) 2006 FORTH: Eleni Orphanoudaki • Portals for information dissemination • Workshops and summer schools: [email protected] and taxonomies for 15 April 2006 for workshops and schools Fraunhofer ICT Group: Michael Krapp classification (20-21 February 2006, later than 15 July 2005 [email protected] Abingdon, UK) FNR: Patrik Hitzelberger • Design of national IST programmes in Events sponsored by ERCIM in 2006: [email protected] FWO/FNRS: Benoît Michel the context of ERA • SOFSEM 2006 - 32nd Conference on [email protected] coordination (9-10 March 2006, Current Trends in Theory and Practice of INRIA: Bernard Hidoine Budapest, Hungary) Computer Science, [email protected] Irish Universities Consortium: Merin, Czech Republic, Ray Walshe Registration for the workshops is free 21-27 January 2006 [email protected] but attendees must pay their own travel • World Wide Web Conference 2006, NTNU: Truls Gjestland [email protected] expenses. Edinburgh, UK, SARIT: Harry Rudin 22-26 May 2006 [email protected] More information: • CAiSE 2006 - 18th Confernce on SICS: Kersti Hedman http://www.cistrana.org/149.htm Advanced Information Systems [email protected] SpaRCIM: Salvador Lucas Engineering, [email protected] Luxembourg, 5-9 June 2006 SZTAKI: Erzsébet Csuhaj-Varjú • CONCUR 2006 - 17th International [email protected] VTT: Pia-Maria Linden-Linna Conference on Concurrency Theory, [email protected] Bonn, Germany, 27-30 August 2006 W3C: Marie-Claire Forgue • DISC 2006 - the International [email protected] Subscription Symposium on Distributed Computing, Subscribe to ERCIM News by: Stockholm Sweden, • sending e-mail to your local editor 19-21 September 2006 • contacting the ERCIM office (see address above) • filling out the form at the ERCIM website at More information: http://www.ercim.org/ http://www.ercim.org/activity/sponsored.html

74 ERCIM News No. 64, January 2006 IN BRIEF

NTNU — The Norwegian University CWI — TT-Medal Project wins ITEA European industry can achieve a reduc- of Science and Technology, NTNU, Achievement Award 2005. The Board tion in test development and test execu- Trondheim in Norway will soon serve of ITEA — Information Technology for tion time and, at the same time, improve as field test laboratory for wireless European Advancement — selected the the product quality. broadband information services. TT-Medal project as the winner of the NTNU has signed a formal cooperative ITEA 2005 Achievement Award. This TT-Medal developed methodologies and agreement with the city of Trondheim, was announced at the 6th ITEA tools based on the TTCN-3 testing lan- the county and the local chamber of guage from the commerce to develop a world-class field European Telecom- laboratory for the testing of mobile and munication Standards nomadic broadband information ser- Institute and introduced vices. Under the agreement, NTNU will them into the European install full wireless coverage for students industry. Eleven partic- and staff on the university’s campus, and ipants took part in TT- will extend the wireless network to cover Medal, including three the commercial areas of downtown Michael Schmidt, Jens Herrman (DaimlerChrysler), Erik ERCIM members: Trondheim. Altena (LogicaCMG), Natalia Ioustinova (SEN2, Centrum voor SEN2 research group Wiskunde en Informatica), Colin Willcock (Project Manager of CWI, Fraunhofer The goal is to preserve session mobility of TT-Medal, Nokia) at the 6th ITEA Symposium, Helsinki, Institute for Open (roaming) throughout the entire network. Finland, 13-14 October, 2005. Communication The coverage in the first stage of the Systems FOKUS and VTT. The results pro- Symposium in Helsinki, Finland, 13-14 vide an opportunity for European test October 2005. Within TT-Medal (Test & tool suppliers and consultants to position Testing Methodologies for Advanced themselves better in a world market that Languages) researchers developed a has been dominated by the USA. generic standardised solution for soft- http://www.cwi.nl/sen2 ware system testing. With the results, http://www.tt-medal.org

CWI — E-Quality expertise centre for INRIA — Objectweb and Orientware Wireless coverage in the Trondheim quality of ICT services. With the fast- signed a memorandum of understand- region. growing number of new ICT services ing to promote the adoption of open- and applications, it is becoming more source middleware in China. Officials difficult to guar- from the Chinese Ministry of Science development will be approximately 5 antee the and Technology (MOST) and INRIA km2, with a target start date of 15 required signed the memorandum on 4 November August, 2006. quality lev- 2005. ObjectWeb is a consortium of els. To leading companies and research organi- The first development phase will employ strengthen zations to produce next generation of Wi-Fi technology, but will gradually research in open source middleware. ObjectWeb shift to other technologies, such as the field of and its Chinese counterpart Orientware, WiMAX, as the technology matures. Quality of announced their decision to build a com- Approximately 20,000 university stu- Service, CWI, TNO Information and mon open source middleware platform dents are expected to comprise the user Communication Technology and the and Orientware agreed to release several community for the planned network dur- University of Twente founded the E- of its middleware components in open ing the first phase in the autumn of 2006. Quality expertise centre in the source. It is widely anticipated that The Trondheim Wireless City pro- Netherlands. The official kick-off took future ubiquitous computing environ- gramme is an effort that welcomes out- place on 30 September 2005 in ments will be highly dynamic, subject to side enterprises, including content and Enschede. Goals of the new centre are to constant changes and of ever-increasing technology providers who would like to transfer knowledge to the market, to fur- complexity. This in turn motivates the test their ideas in the planned field labo- ther expand the knowledge and expertise construction of dynamically config- ratory. of the partners of joint projects and to urable software infrastructures to pro- http://www.ntnu.no/ikt/labs/tt/index_e.htm train specialists. The E-Quality partici- vide a consistent, systematic basis for pants bring in a wealth of expertise in the system evolution, control and manage- area of Quality of Service, each with a ment. different and complementary focus. [email protected]

ERCIM News No. 64, January 2006 75 ERCIM – The European Research Consortium for Informatics and Mathematics is an organisation dedicated to the advancement of European research and development, in information technology and applied mathematics. Its national member institutions aim to foster collaborative work within the European research community and to increase co-operation with European industry.

ERCIM is the European Host of the World Wide Web Consortium.

Austrian Association for Research in IT Institut National de Recherche en Informatique c/o Österreichische Computer Gesellschaft et en Automatique Wollzeile 1-3, A-1010 Wien, Austria B.P. 105, F-78153 Le Chesnay, France Tel: +43 1 512 02 35 0, Fax: +43 1 512 02 35 9 Tel: +33 1 3963 5511, Fax: +33 1 3963 5330 http://www.aarit.at/ http://www.inria.fr/

Council for the Central Laboratory of the Research Councils, Rutherford Appleton Laboratory Norwegian University of Science and Technology Chilton, Didcot, Oxfordshire OX11 0QX, United Kingdom Faculty of Information Technology, Mathematics and Tel: +44 1235 82 1900, Fax: +44 1235 44 5385 Electrical Engineering, N 7491 Trondheim, Norway http://www.cclrc.ac.uk/ Tel: +47 73 59 80 35, Fax: +47 73 59 36 28 http://www.ntnu.no/ Consiglio Nazionale delle Ricerche, ISTI-CNR Area della Ricerca CNR di Pisa, Via G. Moruzzi 1, 56124 Pisa, Italy Spanish Research Consortium for Informatics Tel: +39 050 315 2878, Fax: +39 050 315 2810 and Mathematics c/o Esperanza Marcos, Rey Juan Carlos http://www.isti.cnr.it/ University, C/ Tulipan s/n, 28933-Móstoles, Madrid, Spain, Tel: +34 91 664 74 91, Fax: 34 91 664 74 90 http://www.sparcim.org Czech Research Consortium for Informatics and Mathematics FI MU, Botanicka 68a, CZ-602 00 Brno, Czech Republic Tel: +420 2 688 4669, Fax: +420 2 688 4903 Swedish Institute of Computer Science http://www.utia.cas.cz/CRCIM/home.html Box 1263 SE-164 29 Kista, Sweden Tel: +46 8 633 1500, Fax: +46 8 751 72 30 Centrum voor Wiskunde en Informatica http://www.sics.se/ Kruislaan 413, NL-1098 SJ Amsterdam, The Netherlands Tel: +31 20 592 9333, Fax: +31 20 592 4199 http://www.cwi.nl/ Swiss Association for Research in Information Technology c/o Prof. Dr Alfred Strohmeier, EPFL-IC-LGL, CH-1015 Lausanne, Switzerland Fonds National de la Recherche Tel:+41 21 693 4231, Fax +41 21 693 5079 6, rue Antoine de Saint-Exupéry, B.P. 1777 http://www.sarit.ch/ L-1017 Luxembourg-Kirchberg Tel: +352 26 19 25-1, Fax +352 26 1925 35 http:// www.fnr.lu Magyar Tudományos Akadémia Számítástechnikai és Automatizálási Kutató Intézet FWO FNRS P.O. Box 63, H-1518 Budapest, Hungary Egmontstraat 5 rue d’Egmont 5 Tel: +36 1 279 6000, Fax: + 36 1 466 7503 B-1000 Brussels, Belgium B-1000 Brussels, Belgium http://www.sztaki.hu/ Tel: +32 2 512.9110 Tel: +32 2 504 92 11 http://www.fwo.be/ http://www.fnrs.be/ Irish Universities Consortium Foundation for Research and Technology – Hellas c/o School of Computing, Dublin City University Institute of Computer Science Glasnevin, Dublin 9, Ireland P.O. Box 1385, GR-71110 Heraklion, Crete, Greece Tel: +3531 7005636, Fax: +3531 7005442 Tel: +30 2810 39 16 00, Fax: +30 2810 39 16 01 http://ercim.computing.dcu.ie/ FORTH http://www.ics.forth.gr/

Fraunhofer ICT Group Technical Research Centre of Finland Friedrichstr. 60 P.O. Box 1200 10117 Berlin, Germany FIN-02044 VTT, Finland Tel: +49 30 726 15 66 0, Fax: ++49 30 726 15 66 19 Tel:+358 9 456 6041, Fax :+358 9 456 6027 http://www.iuk.fraunhofer.de http://www.vtt.fi/tte

✄ Order Form I wish to subscribe to the If you wish to subscribe to ERCIM News ❏ printed edtion ❏ online edition (email required) free of charge or if you know of a colleague who would like to Name: receive regular copies of ERCIM News, please fill in this form and we Organisation/Company: will add you/them to the mailing list. Address: Send, fax or email this form to: ERCIM NEWS 2004 route des Lucioles BP 93 Post Code: F-06902 Sophia Antipolis Cedex Fax: +33 4 9238 5011 City: E-mail: [email protected] Country

Data from this form will be held on a computer database. By giving your email address, you allow ERCIM to send you email E-mail:

You can also subscribe to ERCIM News and order back copies by filling out the form at the ERCIM website at http://www.ercim.org/publication/Ercim_News/