InspiredResearch Winter 2019 Issue 15 News from the Department of Computer Science, University of Oxford The Knob attack on Bluetooth Read more – p24

Online radicalisation: Detecting extremist content on Twitter – p17

Bayesian Deep Learning: The work of the Frontier Development Lab – p26

Data Security: Educating users with smartwatch games – p28

compscioxford CompSciOxford DepartmentOfComputerScienceUniversityOfOxford Inspired Research CONTENTS is a twice-yearly newsletter published by the Department of Computer Science at the University P4-12 News of Oxford. If you would like to learn more P13 Farewell to… about anything you read in these pages, please get in touch: P14-16 Student focus [email protected] To subscribe to future issues, e-mail: [email protected] To download previous issues, visit Feature articles www.cs.ox.ac.uk/inspiredresearch P17 Detecting extremist content online

P18 New standards to protect children online

P19 Engaging with your children: recipes for bringing up the ‘AI natives’

P20-21 New research creates computer simulations of Editorial board heart attacks Kiri Walden (Editor) Suvarna Designs (Designer) P22 Virtual Assay and experience with industry Sarah Baldwin P23 Training autonomous vehicles using real-life Leanne Carveth Rosanna Cretney human behaviour Emma Dunlop David Hobbs P24-25 The knob attack on Bluetooth Stefano Gogioso P26-27 Bayesian Deep Learning for all humankind Elisa Passini Helena Webb P28 About time: changing security behaviour through Standa Živný smartwatch games

Contributors p26-27 Sara Dutta, Paula Fiddi, Yarin Gal, Stuart Gillespie, Václav Janeček, Klaudia Krawiecka, Nic Lane, Peter Marinov, Suzanna Marsh, Hector Martinez-Navarro, Mariam Nouh, Alina Petrova, Elisa Passini, Kasper Rasmussen, Blanca Rodriguez, Oliver Samson, Carolyn Ten Holter, James Parr, Helena Webb, Shimon Whiteson, Meredydd Williams, Michael Wooldridge, Jun Zhao, Maaike Zwart.

2 Inspired Research: News from the Department of Computer Science, University of Oxford Letter from the Head of Department

This is my first ‘letter’ since my return from sabbatical, and it’s clear that 2019 has been an exciting year for the department.

We continue to garner awards and recognition across different areas of Computer Science research. Professor Marta Kwiatkowska, Professor Yarin Gal, and alumnus Daniel Nichol have all won major individual awards, while various teams have made an impact at international conferences and in publications of note. It is no fluke that for the second New Starters year running the University of Oxford has been ranked first in the world for Computer Science in The Times Higher Education 2020 World We welcome new members University Rankings. of the department, and some existing colleagues who have 2019 has been a period of change for our community, with many long- changed role. They are (in standing members of the department leaving us. But it has also been a joining/moving order): time of positive development. We now have an interdisciplinary Law and Tracy Marshall, Computer Science course which is run jointly with the Law Faculty. We Nikki Theofanopoulou, are also happy to welcome Oliver Sampson in his role as Industry Liaison Administrator. Mital Kinderkhedia, Yongchao Huang, We continue in our endeavour to widen access to our courses. DeepMind Abhishek Dasgupta, have generously funded four more graduate scholarships, for students Hugo Paquet, identifying as female, BME or from households with traditionally low Alexandra Pendleton, progression to higher education, who want to pursue a full-time master’s Jennifer Lockie, degree in the Department of Computer Science. Professor David Gavaghan and his team ran a Uniq+ a summer school designed to give Gabrielle Alexis, participants the chance to experience what postgraduate study at Oxford Sahar Vahdati, is like. The summer school is for people who might find postgraduate Michaela Walsh, study difficult due to socioeconomic or financial circumstances. Stefano Germano, Qiyi Tang, We are very lucky to attract many talented David Hobbs, applicants each year and our department is so popular that during the Oxford University Patrick Hohenecker, undergraduate open day, for the first time we Robert Newnham, had to use a marquee in the University Parks to Bo Li, accommodate all our visitors. Georgios Birmpas, Sarah Retz-Jones, Once again this newsletter is full of interesting Josephine Francis, items about our department and the work we do. I am looking forward to 2020, which promises to be Hadrien Pouget, another busy year for our ever-growing community. Lei Wang, Ahmad Atamli, Professor Michael Wooldridge Jenny Wang, December 2019 Stefano Gogioso, Anita Hancox, Alina Petrova, DPhil student, Adrian ‘AJ’ Duncan, sadly David Tena Cucala, passed away in October. Adrian worked in Luca Laurenti, the cyber security theme, and was a student Mirco Giacobbe, at Worcester College. He had been unwell for Emanuele La Malfa, some time. He leaves a wife and two daughters and our thoughts are with them, and his many Leema Chapman, friends and colleagues here at Oxford, who will Pan Hu, miss him greatly. Eike Neumann.

www.cs.ox.ac.uk Issue 15 Winter 2019 3 NEWS

Sir Nigel Shadbolt explain and inspire, entertain and News in brief inform. Despite the distemper of judge of Royal the times, and claims we live in a Professor Michael Wooldridge Society book prize post-truth age, there is a real hunger was part of a prestigious line for books that explain our scientific up of guests for the ‘Tech Tonic Professor of Computer Science Sir understanding of world around us. Stage’ at the FT Weekend festival. Nigel Shadbolt chaired a panel of There is huge interest in the triumphs He was a member of a panel five judges to choose the 2019 Royal and tribulations of science and the which discussed the question, Society Insight Investment Science human stories behind the scientific ‘Is Our Future Going To Be One Book Prize. The prize celebrates the endeavour.’ Long Gap Year?’ The panel also very best in popular science writing featured Calum Chace of the from around the world. The winner of the 2019 Prize was Economic Singularity Club and announced at an awards ceremony Kathryn Parsons, founder of Sir Nigel said, ‘It was a privilege to at The Royal Society on Monday Decoded, with Madhumita Murgia be chairing the judging panel for 23rd September, hosted by Professor the FT’s European technology the Royal Society Science Book Brian Cox. Caroline Criado Perez correspondent as moderator. Prize 2019. The prize celebrates was announced as the winner, for great writing about science and the her book Invisible Women: Exposing Sir Nigel Shadbolt was named scientists who make it happen. There Data Bias in a World Designed for one of Business Insider’s UK Tech has never been a more exciting Men. Caroline, a writer, broadcaster 100: The 100 most influential time for popular and accessible and feminist campaigner received a people shaping British technology science writing. There is a huge cheque for £25,000, (and £2,500 was in 2019. appetite for books across all topics awarded to each of the other five Read more: bit.ly/2PkKaRQ of science and writers who can shortlisted authors).

We were delighted to host Dame Wendy Hall (University of Southampton) whom as part of the Department of Computer Science Strachey Lecture series gave a talk this term entitled ‘The Windmill of Your Mind: Reflections on career in Computer Science research’ in the Mathematical Institute, University of Oxford.

Members of the department contributed to the IF-Oxford @SamFairbrother Science and Ideas Festival, presenting two workshops. Mixing music who went on to take a PhD in ‘Managing your Digital Self’ was a Composition from the University visual exploration of the personal with maths of Manchester. The Barbican commission completes a triptych data routinely generated in daily The Barbican launched ‘Life Rewired’, of orchestral works, following Torus life, through smart devices at an arts and learning season, which (Concerto for Orchestra), which home, digital mobile devices used ran throughout 2019 exploring what won the orchestral category of the by millions of families, as well as it means to be human in the face of 2017 British Composer Awards, various portable devices which technological and scientific forces and sphere, which was premiered surround us in everyday life. In that are dizzying in their speed, scale by the Bamberger Symphoniker in the ‘Making Faces’ workshop and complexity. As part of this season the same year. Emily is the Director Niki Trigoni talked to members Sir Simon Rattle opened the London of PRiSM, the Centre for Practice & of the public about the latest in Symphony Orchestra’s 2019/20 Research in Science & Music at the ‘multimodal recognition’, which season on 14 Sep 2019 with the Royal Northern College of Music in uses facial recognition technology world premiere of Antisphere, a new Manchester, which brings together work by composer Emily Howard. creative collaborations between in combination with voice and the sciences and music including walking gait analysis to help Emily is an Oxford graduate in cutting-edge research involving identify individuals. Mathematics and Computer Science AI-assisted composition.

4 Inspired Research: News from the Department of Computer Science, University of Oxford NEWS European Research Council Starting Grant awarded to Nic Lane Nic Lane has been awarded a energy) that are currently slowing to deep learning efficiency, a prestigious ERC Starting Grant for the advancement of the field, and desperately needed step given his project ‘REDIAL: Re-thinking limiting full global participation in current over reliance on empirical Efficiency in Deep Learning under this research to only the largest of observations. Accelerators and commodity and companies. processors’. Second, it aims to develop new The goal of REDIAL is to solve core architectures and methods for Over a short period of time, technical challenges that span the training and inference that tackle breakthroughs from the field of deep areas of machine learning and system core efficiency bottlenecks, such learning have transformed how research which collectively can as: dependencies preventing computers perform a wide-variety enable a radical jump in the efficiency parallelisation and excessive on-chip of tasks such as recognising a face, of deep learning. It aims to address data movement; while also opening driving a car or translating between both the challenge of high training new opportunities including the languages. costs and time, as well as the barrier greater adoption of analog processing to deploying models on constrained within accelerators. REDIAL aims Not only has deep learning become devices (like wearables, sensors) to change the way the world trains an everyday tool, it is also the most that currently require new efficiency its models, and deploys them to promising direction for tackling techniques to be invented each time constrained devices, by producing a number of still open problems a deep learning innovation occurs. a series of new deep architectures in machine learning and artificial and algorithms with properties intelligence. However, routine deep To accomplish this REDIAL takes that promote high efficiency that learning activities (such as training two complementary approaches. can serve as a foundation for new a model) exert severe resource First, it seeks to build a theoretical machine learning innovation. demands (eg memory, compute, understanding of current approaches

Daniel Nichol wins The committee who chose to highly prestigious journals and to a make this award commented, few more papers under publication.’ Reinhart Heinrich ‘Daniel Nichol’s thesis is likely to Award become a landmark in an issue of Dan’s supervisors were Professor critical significance. It studies and Peter Jeavons (Oxford, Computer Recent Computer Science graduate proposes strategies to deal with Science) and Professor Alexander Daniel Nichol has won the 2019 drug resistance in the treatment of R. A. Anderson (Moffitt, Integrated Reinhart Heinrich Award for his cancers and infectious diseases Mathematical Oncology). doctoral thesis ‘Understanding Drug taking into account the genotype- Resistance through Computational phenotype mapping. New Models of the Genotype–Phenotype mathematical models are used to Mapping’. The award is made by the better understand the underlying European Society for Mathematical mapping mechanisms, integrating and Theoretical Biology (ESMTB) genetics, environmental signals and and will be formally given in a stochasticity. The thesis gave rise ceremony in Heidelberg in 2020. to three papers already published in

Samsung invests in fast-growing field of semantic Motik, Ian Horrocks and Bernardo technology. The investment Cuenca Grau. The company’s spinout company round was also supported by the technology turns an organisation’s University of Oxford and Oxford data into machine-readable Oxford Semantic Sciences Innovation PLC. knowledge, combining diverse data Technologies sources and then enabling users to Spun out of the Department of query them with powerful reasoning. Samsung Ventures has led an Computer Science in 2017, Oxford investment round that raised over Semantics is developing and Samsung’s investment will enable £3 million for Oxford Semantic commercialising RDFox, the high- Oxford Semantics to bring its Technologies Limited, an Oxford performance knowledge graph powerful data integration and University spinout company and semantic reasoning engine reasoning engine to more clients. founded by pioneers in the developed by Professors Boris

www.cs.ox.ac.uk Issue 15 Winter 2019 5 NEWS News in brief Oxford ranked University of Cambridge and first in world for Stanford University. The ranking is based on criteria David Rogers is an external Computer Science measuring teaching, research, lecturer who has been awarded industry income, international MBE for services to cyber security. For the second year running the outlook and citations, which are David taught a module on Mobile University of Oxford has been combined to provide a comparison Systems Security for our part-time ranked first in the world for of universities worldwide. Since professional MSc in Software & Computer Science in The Times 2017 Oxford has also annually Systems Security and is himself a Higher Education 2020 World been ranked the top university in graduate of our MSc programme. University Rankings. This is ahead the world according to the same of California Institute of Technology, publication. Andrew Markham has featured as an Oxford Sparks scientist in the spotlight. More details on Oxford brings together experts and how Andrew became a Computer Scientist and Engineer here: students from Computer Science and Law oxfordsparks.ox.ac.uk/content/ andrew adapted from an article by Václav Janeček

In July St Edmund Hall hosted a On Tuesday 15th October, the solo exhibition entitled Oxford: University of Oxford launched its new An AI-art View’ showcasing interdisciplinary course, Law and artificial-intelligence created digital Computer Science. paintings and video art pieces. Developed by Mayur Saxena This course, jointly offered by the (2018, MSc Computer Science), Department of Computer Science the AI-agent algorithmically (co-convenor Professor Tom Melham) captures the essence of iconic and the Law Faculty (co-convenor Oxford landmarks from images Professor Rebecca Williams), will on social media then reproduces introduce students from both backgrounds to the terrain at the boundaries these as Impressionistic digital of their two disciplines. The overarching theme is understanding Law and artworks. Computer Science at their interface.

Katie Dicks has been awarded a Such interdisciplinary understanding requires both lawyers and computer certificate for her participation on scientists to develop an appreciation of the way in which they typically the OBEA (Oxfordshire Business approach problems, with very different analytic tools. A key pedagogical Education Alliance). The group strategy for the course is to teach Law and Computer Science students work with students from four local together, and in particular for them to collaborate on a group-work practical state comprehensive schools on a exercise. This will accelerate both groups’ acculturation to each other’s project. The project has involved analytic perspectives through learning from each other as well as from with working with members of the faculty. department of Computer Science to work together on solving a The opening lecture as well as the lively follow-on discussion was led by problem. Professor Richard Susskind, a OBE former Oxford student, who himself has combined the two disciplines since late 1980s and is now considered one Professor David Gavaghan the leading figures in the field. and project team ran UNIQ + a summer school designed to The research behind this course is part of an ongoing Oxford project titled give the chance to experience ‘Unlocking the Potential of Artificial Intelligence for English Law’ which what postgraduate study is like is run by researchers in the Oxford departments and faculties of Law, at the University of Oxford, for Economics, Computer Science, Education and the Saïd Business School, talented individuals who would and led by Professor John Armour of the Faculty of Law. find continuing into postgraduate study a challenge because of socioeconomic or financial As a global leader in education, Oxford University is committed to sharing circumstances. knowledge and insights related to the course and will release multiple open Read more: ox.ac.uk/uniqplus access materials as the course progresses.

6 Inspired Research: News from the Department of Computer Science, University of Oxford NEWS

DeepMind renews open to applicants identifying as Science, says: ‘We are delighted female who are ordinarily resident in that DeepMind has renewed their commitment to the UK. scholarship programme for under- represented groups. Our goal is to under-represented Vice-Chancellor, Professor Louise deliver the best possible education to students at Oxford Richardson, says: ‘We are very the best students, irrespective of their grateful for DeepMind’s continuing background. This scheme is a huge DeepMind, the leading British artificial support of underrepresented groups boost for us in that mission.’ intelligence company, has renewed at Oxford. This supports our strategy and extended its commitment to to increase the diversity of students Professor Ronald Roy, Head of the supporting students at the University and the diversity of perspectives in Department of Engineering Science, of Oxford following the success of the University.’ adds: ‘The Department is very its inaugural scholarships for under- much appreciative of DeepMind’s represented groups, announced in Dr Demis Hassabis FRS, co-founder commitment to supporting 2018. and CEO of DeepMind, says: ‘AI has postgraduate education in information incredible transformative potential, engineering. Our goal to engage more Four more DeepMind graduate but if we’re going to develop a women in this rapidly emerging field scholarships for students wishing technology that benefits everyone, will be significantly advanced by the to pursue a master’s degree in the it is critical that the people working company’s generosity.’ Department of Computer Science on it are representative of the wider will be made available for students world. That’s why the next generation DeepMind was established in London commencing courses in 2020–21. of researchers, from all backgrounds, in 2010 and is now a world leader in These are open to individuals should feel able to access the best artificial intelligence research and its ordinarily resident in the UK who educational opportunities. We’re application for positive impact. The belong to one or more of the following proud to be working with leading DeepMind Scholarships are part of groups: identifying as female, BME or universities, such as the University a wider initiative by the company to from households with traditionally low of Oxford, to offer talented students broaden participation in science and progression to higher education. from under-represented backgrounds support wider research in the UK. a chance to reach their full potential Three new DPhil scholarships, through scholarships and other More details of the DeepMind with a preference for students of programmes, collectively broadening Scholarships will be available soon on machine learning, in the Department access to science and research.’ https://deepmind.com/blog. of Engineering Science will also be launched thanks to a further gift from Professor Michael Wooldridge, Head To read more please visit: https://bit. DeepMind. These scholarships are of the Department of Computer ly/33aFGAy Yarin Gal announced as Turing AI Fellow

Yarin Gal is one of five new Turing challenges, new safe and robust AI Fellows announced by The Alan AI tools will be developed for Turing Institute. The Turing AI Fellows, responsible use in industry. who have been appointed for five years, are drawn from a wide variety Adrian Smith, Institute Director and of disciplines and backgrounds and Chief Executive of The Alan Turing will be tackling research challenges Institute, said: ‘We look forward ranging from sustainable aviation to what this talented group of to AI for discovery in data intensive Turing AI Fellows will bring to our astrophysics. vibrant research community and we welcome their contributions Yarin will work on democratising to our growing Institute. There is safe and robust AI. While already in vast potential for their diverse work use in industry and academia, major to be transformative in both the obstacles still stand in the way of foundations and applications of AI deploying deep learning AI safely and I am confident they will push and responsibly. Yarin proposes to the boundaries of what these new tackle these problems by building technologies can do for the good of community challenges derived society.’ from real-world applications of AI in industry. With the community To read more: https://bit.ly/34Sxwyv competing on these public

www.cs.ox.ac.uk Issue 15 Winter 2019 7 NEWS Marta Kwiatkowska named as BCS Lovelace Medal Winner 2019

Professor Marta Kwiatkowska has developing modelling and automated been named as the recipient of the verification techniques that can 2019 BCS Lovelace Medal, the guarantee the stable, safe, secure, top award in computing in the UK, timely, reliable and resource-efficient awarded by BCS, The Chartered operation of computing systems.’ Institute for IT. PRISM is widely used for research The award is presented annually to and teaching across the globe; achievements. The impact she has individuals who, in the opinion of the it is highly cited, and has been made to Computer Science, and her BCS Academy Awards Committee, downloaded over 70,000 times contribution to academic research, have made a significant contribution and used in over 500 publications makes her a richly deserving recipient to the advancement of Information to date. It has been shown to be of this prestigious award.’ Systems. effective for modelling a wide range of applications including Marta says of being presented with Marta has been recognised for her communications protocols, molecular the award: ‘I’m incredibly honoured major contributions to probabilistic and cell biology, and more recently, to be receiving the 2019 Lovelace and quantitative verification. She has adaptive and Artificial Intelligence Medal. It’s really exciting to have made significant contributions across systems. It has proved its usefulness my work recognised via the award the breadth of theory, applications in many distinct fields, including and emphasises the importance and and software tools. Her research and distributed and cloud computing, potential of research in probabilistic her software system PRISM have wireless networks, security, robotics, and quantitative verification. Over the made a huge impact on Computer quantum computing, game theory, years, I have worked in conjunction Science in the UK and worldwide. biology and nanotechnology. with many collaborators and am Since 2001 she has led the grateful for their contributions, none development of the highly influential The award will be presented to Marta more so than Dr Gethin Norman and probabilistic model checker PRISM. at a ceremony at the Royal Society in Professor David Parker, who worked London next spring. on the first implementation of PRISM Marta explains: ‘Computing and have provided great support ever infrastructure has become Steve Furber, Chair of BCS since.’ indispensable in our society, with Academy Awards Committee says: examples ranging from online ‘Professor Kwiatkowska has made Nominations for the 2020 award open banking, to intelligent vehicles and an outstanding contribution to the on Monday 21 October 2019 and electronic medical equipment. understanding and advancement of close on Friday 24 January 2020. Software faults in such systems can computing and I am delighted that we have disastrous consequences. are awarding her the Lovelace Medal More information about the award My research is concerned with in recognition of her wide range of can be found at www.bcs.org.uk Oxford Hack 2019

Oxford Hack 2019 ran again this year at the University of Oxford on 16 – 17 November 2019. It is an event designed to encourage student teams to build something brilliant in 24 hours. Among the group were a number of Department of Computer Science teams including some building: a ‘distributed multi-agent reinforcement learning’ system for web browsers (https://bit.ly/33bWEOR); a machine-learning android app that lets you have conversations with yourself (https://bit.ly/2pFw6YT); and a satirical General Election twitter bot using Tensorflow (https://bit.ly/2O8yGjC). The hackathon had 19 sponsored awards such ‘most entertaining hack’ and Photo above: (left to right) ‘most stylish hack’ which were given to various teams and their projects. Deniz De Barros, Edd Salkield, Alexandra Manciu, Șerban Șlincu, Josh Smailes, Further information on the awards and winning hacks available here: Sam Boyer. https://bit.ly/2rdeT9D

8 Inspired Research: News from the Department of Computer Science, University of Oxford NEWS 100+ Brilliant Women Oxford Computer Science Conference in AI & Ethics award winners The 2019 Oxford Computer Science Conference took place on 6 June and was a great success. The conference featured talks and poster sessions, and concluded with an award ceremony and conference dinner.

The winners of the Oxford The ORBIT project has created The conference was always intended Computer Science Conference Oxford’s first carbon-neutral to be carbon-neutral, and every 2019 Awards are: conference. The sold-out 100+ decision was taken with a view to 1st presentation award Brilliant Women in AI & Ethics took minimising the carbon footprint. Ulrik Lyngs place in September and focused on The carbon that was created is 2nd presentation award three key themes - the climate crisis being offset with tree-restoration Sean Sirur threatening animal and human life; projects in the northern hemisphere the pushback against untrammelled and rainforest-protection in the 1st poster award AI; and the need for diversity in south, ensuring that the event was Lingyi Yang STEM disciplines. accredited as carbon-neutral by an 2nd poster award assessment company. Isaac Dunn The conference included speeches 1st abstract award from Baroness Beeban Kidron; The breakout workshops, each Ben Fernando Angela McKane; Jacquelyn Krones; chaired by an expert in the field, were 2nd abstract award Beena Ammanath and Dame the highlight of the day. Conference Emma Bluemke Wendy Hall. A prevalent theme delegates brought their insight to bear was the competitive advantage to on four key topics, ensuring that all be obtained in AI through greater those who attended the conference AI@Oxford diversity, and the need for ethical contributed to its outputs. considerations to ensure that conference the impact of tech on society is The incredibly positive response The AI@Oxford conference took more positive. Offering a robust to the event demonstrated the place in September and many perspective was author Jeanette importance and international impact academics from our department Winterson, who has spoken out of the work undertaken by ORBIT were involved. On the first day of before on her belief that AI will have and the relevance of responsible a packed two-day schedule Mike a disproportionate negative effect research and innovation to cutting- Wooldridge gave the opening talk on women. The day was enriched edge technologies. ORBIT is ‘The World of AI’, Daniel Kroening by a panel of A-level students creating a report from the data spoke about ‘The Augmented World’, who had entered a competition to produced on the day that will be Jun Zhao contributed to a session on attend - they shared their thoughts released later this year. ‘Impact of Trust in AI’, Alex Rogers on artificial intelligence as well as was a speaker in a session on ‘Earth receiving their prizes. Read more: www.orbit-rri.org/ Systems and AI’ and Marina Jirotka spoke about ‘Responsible Research and Innovation’. Eighteen Oxford Computer Science On day two, Sir Nigel Shadbolt papers at NeurIPS-2019 gave a talk on Ethics in AI, then Niki Trigoni, Alex Rogers and Andrew Members of the Department of Computer Science have co-authored Markham all spoke as part of a 18 papers accepted for the forthcoming NeurIPS conference to session on ‘Cyber Physical Systems’, be held in Vancouver, Canada, in December 2019. NeurIPS is the Thomas Lukasiewicz contributed premier international forum for research on machine learning: this year to the session on ‘AI and Legal 6743 papers were submitted, of which only 1428 were accepted (an Reasoning’ and the whole conference acceptance rate of just 21%). wrapped up with a ‘Future Makers Question Time Special’ panel discussion which included Peter For more information on NeurIPs-2019 see: nips.cc/Conferences/2019/ Millican (chair) and Mike Wooldridge.

www.cs.ox.ac.uk Issue 15 Winter 2019 9 NEWS Oxford Women in Computer Science

Oxford Women in Computer Science actively supports and promotes women in the field of Computer Science, running social and academic events to provide networking opportunities and a support network for students. OxWoCS events are inclusive and opened to everyone in CogX 2019 – The Festival of AI and to management consulting by Oliver the university. Emerging Technologies. All those Wyman. who took part had a great time; it Earlier this year we continued our was amazing to see the opportunities OxWoCS Distinguished Seminar [Pictured above] The 2019-20 within the industry and networking OxWoCS Committee Members Series with joint lectures with the with professionals. Top (left to right): Yuge Shi (Sponsorship Department of Computer Science. Officer), Jasmine Rienecker (Conference Talks included those by Julie McCann We are kick starting the 2019-20 Coordinator), Stacy Xingyue Pu (Seminar (Imperial College), Mary Wootters academic year with our annual Series Coordinator), Eleonora Giunchiglia (Stanford Uni), Jenna Wiens (Uni. Of (Conference Coordinator), Marilena Bescuca welcome event and a tech talk Bottom (left to right) Michigan) and Olga Sorkine-Hornung (Secretary) : Clara by our sponsor, Google, on ‘The Pavillet (Industry Events Officer), Tajwar Nasir (ETH Zurich). We will continue with Art of Clean Code.’ These will be (Outreach Officer), Paula Fiddi (President), more tech talks on various areas followed by events and activities Klaudia Krawiecka (Outreach Officer), Lonie including but not limited to mobile organised by our other sponsors: Sebagh (IT Officer) Committee members not in picture: systems, computer vision, machine office visits to QuantumBlack; Catherine Tong (Industry learning and artificial intelligence. We Events Officer), Romy Minko (Researchers workshops on algorithm fairness and Rep), Annazita Barry and Teodora Musatoiu have made trips to sponsors HQs, machine learning by QuantumBlack; (Undergrad Rep), Asmita Poddar (MSc./ visited GlaxoSmithKline and attended blockchain technology by Zilliqa Postgrad Rep), Lucy McEvoy and Eleanor a QuantumBlack sponsored event, Blockchain; and applying tech skills Williams (Social Events Coordinators)

A trip to GHC 2019 Klaudia said: ‘The key theme this year in technology across the world. I was “We will. I will.”, and my personal feel refreshed and ready to face new The Department of Computer motos are: we will keep supporting challenges. The experience was Science and OxWoCS co-funded one another and empowering one empowering and inspiring and I look three scholarships to attend this another. I will continue to be an forward to building and maintaining year’s Grace Hopper Celebration advocate for young children and connections I made during the (GHC). GHC is the world’s largest educate them about technology.’ event. My quote for this year’s theme conference of women technologists is: “I will actively share women’s gathering students, academics and Alina said: ‘The conference is a contributions in tech and create professionals across a vast range of perfect demonstration of how truly technology to improve lives.”’ specialties and areas of expertise. diverse the area of Computer Science is. It was not only a huge inspiration DPhil students Klaudia Krawiecka, for me, but also a great opportunity to Paula Fiddi, and Alina Petrova improve my networking skills, as I am attended the 2019 GHC in Orlando, transitioning from being a student to Florida. It was the largest GHC being a postdoc.’ event to date, with more than 25,000 attendees, over 250 industry Paula said: ‘The Grace Hopper sponsors, and hundreds of talks, Conference was filled with such seminars, and workshops. amazing energy from 25,000+ women

10 Inspired Research: News from the Department of Computer Science, University of Oxford NEWS OxWoCS collaborates with Zilliqa to bring diversity to the forefront of Blockchain Zilliqa, a high-throughput public business and technological aspects and Oliver Wyman as OxWoCS blockchain platform, has launched of implementation. Taking place sponsors. the return of Blockchain A-Z, a over the course of four weeks, series of blockchain education these workshops aim to provide a Paula Fiddi, OxWoCS President workshops, this time with the well-rounded overview of concepts, said, ‘At OxWoCS, our goal is to University of Oxford. Blockchain including philosophical foundations; ensure that female scientists are A-Z aims to encourage greater technical topics; industry insights; presented with equal opportunities gender inclusion from lecture halls and business advisory. As part of to engage with various areas of the and beyond. The workshops have the Demo Day, winning participants tech industry. From established been launched in collaboration were invited to submit their fields of research to nascent with Oxford Women in Computer solutions to Zilliqa’s US$5 million disciplines ripe for exploration, Science Society (OxWoCS) and Ecosystem Grant Programme, in it’s important that women are were open to all students currently the hopes of receiving mentorship, adequately represented in order to enrolled at the University of Oxford. technical tutorship, and a guarantee champion diversity both in action The workshops ran from 31 October of funding for 0% equity in return. and in practice. With Zilliqa’s 2019 and culminated with a Demo emphasis on nurturing growth from Day on November 21, 2019. OxWoCS was founded in 2013, the ground up, this collaboration with the mission of supporting serves to prepare our students with The interactive workshops aim to and empowering female students much-needed support, guidance, equip students with a foundational enrolled in programmes. Zilliqa is and industry connections, to truly understanding of blockchain joined by Google, QuantumBlack, excel in the blockchain space.’ fundamentals from both the Facebook, G-Research, Semmle Industry Liaison Administrator: Oliver Sampson

I have recently been appointed further opportunities for collaboration. greater Industry Liaison Administrator My role also incorporates our success, and for the Department of Computer students, and I believe better we are already Science in order to provide a industry interaction will result in a seeing benefit link and initial point of contact wider understanding of where their from a number between industry and academia. My career can take them. We currently of projects with background is in the life sciences have industrial representation from industrial links. having studied for a DPhil in finance, automotive, pharmaceutical Radiobiology at the University of and telecommunication sectors, each As an initial project, the Oxford and prior to that working offering a different perspective on industry relations section of the for GSK and RAFT in analytical how our graduates can contribute to departmental website has been chemistry and molecular biology their business. We aim to increase given a facelift and now contains positions respectively. Drawing this list in the coming years. This is much more information about the together my experience in both important for the advancement of the ways industry can partner with us. industrial and academic sectors I department, the career prospects of Please visit www.cs.ox.ac.uk and hope to use this as a foundation to our students, and also in attaining click on industry relations under foster links between the two sectors the best outcome as part of the the innovation tab. Of course the here in Department of Computer Research Excellence Framework list of ways to interact with the Science, University of Oxford. review in 2021. department is not finite and I am always open to new suggestions As Industry liaison administrator I In my new role I am in a position to on how we can further develop have the broader interests of the support nascent projects where there links from inside the department academic community in focus. We is a mutual benefit to the Department or through our industrial partners. are not merely seeking to gain funds and where industry funding and/or to undertake research projects but assets will provide synergistic benefit. Please feel free to stop by my rather to learn and benefit from what Typically, time and administration office in room 104 in the Wolfson industrial partners can bring to the involved in forming these relations is building, Department of Computer equation. For example this might inhibitory, so being able to dedicate Science for a chat or email be data resources, knowledge or effort to this process should provide [email protected]

www.cs.ox.ac.uk Issue 15 Winter 2019 11 NEWS

The Royal Institution Computer Science Masterclasses

Generously sponsored by highlight the links between this the University of Oxford. Topics Morgan Stanley wide-reaching discipline and other run previously have included: subjects. The series of 6 Saturday computing meets biology, machine The Royal Institution morning workshop each lasting learning, sphero robots workshop, Masterclasses are a series of 2.5 hours were lead each week by and turtle graphics. hands-on and interactive a different expert speaker offering extracurricular activities that are students in-depth investigations This year the Department of designed to encourage, inspire of topics in Computer Science, Computer Science welcomed and engage young people in combining theory with interactive back our 24 x Y10 students, who the creativity and practice of exploration working with schools that attended in 2018 as Y9s, and also Computer Science, helping to otherwise do not often engage with welcomed a new cohort of 22 x Y9 students. Between them the two cohorts represent 24 different state schools with up to two students from each school from within our catchment area of state schools in Oxfordshire, Berkshire, Wiltshire, Gloucestershire and Buckinghamshire. We hope to run the programme for a total of three years per cohort and aim to offer these masterclasses in Computer Science again in 2020.

Further information available: bit.ly/2Jk6TJS Open day in the Park

The department regularly opens its courses offered by the Department doors as a part of the university- of Computer Science and as a result wide open day(s) during the year. the number of people attending An ideal opportunity for prospective our open days has increased students to explore Oxford, find out significantly. The most recent more about courses and colleges, as open day took place on Friday 20 well as speak to tutors and students. September and to accommodate the We have had growing interest in increasing numbers the programme the last few years in relation to was split between the Wolfson undergraduate Computer Science Building and a marquee in University the large number of people involved Parks. Throughout the day we had giving talks, demonstrating projects, just over 500 prospective students running social media, and mingling and guests visit the Department of with prospective students. Computer Science. In comparison Thank you to all members of the we had only 123 prospective department involved in the open students and guests in September day. We will be taking part in the 2013 whilst if we go back 10 years next university-wide open days on to September 2009 we only had 48 1 July, 2 July and 18 September prospective students attending the 2020. open day! For further information on open The September 2019 open day days or other outreach events could not have gone ahead without please see: cs.ox.ac.uk/opendays

12 Inspired Research: News from the Department of Computer Science, University of Oxford Farewell to… The Department of Computer Science is a complex community of people. Some are only here for a short time on their way up the career ladder, but a few have been an integral part of the team for years, even decades. This year we are saying a fond farewell to several people who have help to shape this dynamic department as it has grown and evolved throughout the years.

Sharon Backer-Holst launched, so it is fitting that we are Maureen York ‘losing’ her to department spin out Sharon’s official job company Navenio. Another stalwart of title was ‘Department the department, after Administrator / Research There is not a person here who six years as Centre Facilitator / Project doesn’t know Sharon, and we will all Administrator for Manager’ but even miss her hugely. the Cyber Security that complicated title CDT and over 25 doesn’t really get close Julie Sheppard years working in to describing what she did, and the the department, huge contribution she made to the When we celebrated Maureen York has departed for an department over the years. our 60th anniversary exciting new life in Spain. In the in 2017 it was time time she was with us Maureen was Sharon worked tirelessly to find this to look back over particularly involved in the creation ever-growing community more space, the history of our of several Centres of Doctoral using imagination and determination department, and one Training (CDT’s). This included the to somehow create new space in our person was able to tell us all there most recent project she worked existing buildings, and acquiring new was to know. Graduate Studies on, helping Jim Davies set up the buildings when she could. Administrator Julie Sheppard joined new CDT in Health Data Science. the department in 1984, and retired Maureen was great at bringing these In her role she made sure that the this summer as our longest serving complex projects together and we academic and support staff worked member of support staff. will miss both her negotiating skills together, helping us maintain a sense and unique purple hair! of community as we become a far During her career Julie held a number bigger department than the one she of different jobs, most of them David Hobbs, Centre Administrator originally joined. involving student support. As a result for the CDT in Cyber Security many students, past and present comments, ‘Maureen’s calm and Our Head of Department Mike have very fond memories of her. professional attitude, effervescent Wooldridge comments, ‘It would personality and deeply caring be hard to overstate the role that Former student Joe Loughry perspective on life in addition to a Sharon Backer-Holst (formerly Lloyd) commented that, ‘Julie Sheppard love of everything purple, will be has played in the growth of the was the lighthouse I navigated by keenly missed. We have no doubt department over the past decade. throughout my DPhil. I owe her.’ that Maureen will make a massive The mission to grow and broaden positive impact helping others with the department would simply have Laura Jones her new business.’ been impossible without such a dedicated, focussed, and determined Laura was an HR Kelly Ryan administrator and financial manager. Manager who really Oxford is a complex university, and cared about the Finally, Kelly Ryan making change happen here is community she was left earlier in the year. not easy. Sharon transformed the part of. Apart the considerable job of Kelly was our Senior finances of the department, and has overseeing the constant comings and Research Facilitator, fought for us endlessly to get people, goings of staff in a busy academic and instrumental space, and resources. I could not department, she also introduced in the success begin to count the number of times I many initiatives to look after our we have had (and have relied on her utterly. Everybody physical and mental wellbeing. continue to have) in applying for and working in the department today is in being awarded research funding. her debt.’ Laura was also a key member of the She became the ‘Oxford-Turing Athena Swan team, coordinating our University Liaison Manager’ before Sharon has always had a keen award submissions and making sure leaving us to work full time at The interest in the research undertaken that we continue to strive towards Alan Turing Institute in London. by our academics, and especially in better opportunities for women in our the various spin outs that have been department.

www.cs.ox.ac.uk Issue 15 Winter 2019 13 Lego lunch

We have launched a regular Lego lunch hour. Every Wednesday in the Atrium a selection of kits and loose bricks are available so that everyone in the department can spend some time out, doing something relaxing and fun together. The idea of adults building with Lego to relax is not a new one, much has been written about how effective it can be to help take the stress out of everyday life. We welcome all members of the department, and hope to grow the Lego collection as more people get involved. Sport at Computer Science There tends to be a popular idea of what a Computer Scientists might be like, and ‘sporty’ is not the first description that comes to mind. But in this department we have (and have had) students involved in a wide variety of University sports, from competition-level rowing to….Quidditch!

Peter Marinov [pictured below] is I love tennis as it requires me to couples and with a group routine. a Doctoral student working within concentrate and be present in the Basically we’re a crazy bunch the computational biology research moment, as well as to think on my having fun while climbing each theme. He is involved in a number feet. I play for the Uni 4th men’s other, laughing at our fails and of different sports, and says, ‘I’m a team and we won varsity last year. celebrating successes. It makes for tennis player, volleyball player and some very good holiday pictures windsurfer. I play volleyball for more or less the too! same reason but also really love the team goal component, where the team has to play well for us to win a match. I used to play for the Blues and now play for the 2nd Uni team.

I like windsurfing because its relaxing and frees up my mind. I like how unpredictable nature can be and the challenge to ride it.’

Maaike Zwart [pictured right with her dance partner] is a Doctoral student working on research in quantum computing. She is involved in an acrobatic dance group.

She says, ‘I’m part of Flight Club, an acrobatic partner dance club. At Flight Club, we mix several partner dance styles (modern jive, jive, rock and roll, etc) with jaw-dropping acrobatics. We take part in lots of competitions, both as individual

14 Inspired Research: News from the Department of Computer Science, University of Oxford Tokyo exchange

The Department of Computer ‘I visited Tokyo on a scholarship fantastic experience as I was able Science has for many years had in 2008, towards the end of my to observe local life up close. an arrangement with the National second year of PhD. Japanese people are generally very Institute of Informatics (NII) in helpful. It is not difficult to make Tokyo, whereby they will provide In my case, the area of research I friends if one tries. scholarships for our MSc and DPhil collaborated on while I was at NII students to spend time as an intern in was very closely related to my PhD My NII scholarship has significantly their lab in Tokyo. research, which was programming contributed to my research, DPhil languages. and beyond. A chapter of my DPhil NII is a government-sponsored thesis is based on my work done research institute, located right in the Academically it was not very during the internship, and the centre of Tokyo, just next door to the different to working at Oxford. We academic connections fostered Imperial Palace. They have research were all focused on doing good there have been long lasting. I have groups in a variety of areas. In research and trying to publish been constantly collaborating with particular, there is a very strong group good papers. Generally speaking, Japanese researchers since then. under Ichiro Hasuo on categorical Japanese research labs tend to As we speak, a PhD student I got to semantics and co-algebras; they be more formally organised, with know then (who is now an associate also have groups in quantum a professor on top, followed by an professor in a top Japanese computing, cyber security, cyber- associate professor assisted by university) is visiting me at Bristol. physical systems, graph algorithms, an assistant professor. There are verification, and social technology. typically a number of postdocs, I would definitely recommend this PhDs, and MSc students. The scholarship to current students. With so many research areas in groups tend to be very close knit, Visiting other research groups is common, a collaboration between with regular social events that always a good way to generate new our department and NII has many everyone attend. ideas, and Japan is a place where advantages. Jeremy Gibbons is a top-rate research is done. The Visiting Professor at NII and one of Life in Japan is (obviously) very social life is very colourful too.’ his former DPhil students relates his different from that in the UK. So experience as a scholarship intern there is much to explore and learn. Read more about NII here: at NII: Through a Japanese friend I got to www.nii.ac.jp/en/research/centers/ know back in Oxford, I managed Meng Wang (now a lecturer at to do a house-share with three Bristol University) recalls: Japanese young men. It was a

www.cs.ox.ac.uk Issue 15 Winter 2019 15 Student Profile

Sara Dutta : A DPhil in Computational Biology who has gone on to work with the FDA and pharmaceutical and medical device companies.

What course did you study here and when? process to improve DPhil in Computational Biology (DTC Systems Biology prediction of lethal 2009) 2010-2014 arrhythmias. I worked there for 3 years What was your background before that? and then moved to I studied Computer Science BEng at Imperial College Boston over a year ago to work for Dassault Systèmes London and I did my high school at the Lycée François (the 3DEXPERIENCE Platform company) where I Premier in Fontainebleau (France) where I grew up. help pharmaceutical and medical device companies embrace digital transformation throughout their What attracted you to studying Computer Science lifecycle, one of the main ways being through modelling as a subject? and simulation. I always enjoyed mathematics and logic at school and I was fascinated by how it could be applied in Computer How has the course you studied here helped you Science to build algorithms that could solve amazing in your current profession? problems. The work I did at Oxford has really helped lay strong foundations for all the work I have done since then. All What aspects of the course you studied here did the technical knowledge I learnt during my PhD has you particularly enjoy? been directly applied during my research work at the I particularly enjoyed interacting and learning from my FDA and now I use that technical expertise to help colleagues in the department, such as during the student demonstrate the business value of such tools to our life departmental conference and having the opportunity to sciences companies. I also have kept close ties with be a demonstrator and tutor during some undergraduate my team from Oxford throughout and they have always courses. Most importantly my team under Professor been a great source of support throughout my time at Blanca Rodriguez was a great source of support, fun and the FDA and in my current job. knowledge. What advice would you give to current students What did you enjoy about life at Oxford? on applying their knowledge in the workplace, I really valued interacting with friends and peers from when they leave university? other subjects. Balliol MCR dinner parties were a source Computer Science gives you a great set of analytical of exchange and fun as we were a diverse group of and logical skills that can be applied to virtually any PhD students from a range of subjects from Physics, field. Don’t be scared to try areas a little out of your Computer Science, Philosophy, Literature and Economics. comfort zone or that seem a little too ambitious. I also enjoyed that there were always fascinating talks, My 18 year old self would have never thought that conferences, concerts and other events that were easily computational modeling of the heart existed and was accessible. At times it was intimidating, but mostly I was something she would make a career out of. just fascinated by all professors and students around me, and tried to learn from them as much as possible. I have If we went back in time and asked, what would particularly fond memories of being part of the organising the student you have thought about what you are committee for the Balliol’s 750th Commemoration Ball and currently doing – would you have been surprised, rowing for Balliol during Torpids and Summer Eights. proud, amazed? I think I would have been impressed and I would What did you do when you left Oxford? have found it hard to believe that there are so many I went to work for the U.S. Food and Drug Administration opportunities to continue to learn and apply my in the Washington D.C. (to be precise, Silver Spring MD) knowledge outside of the traditional academic route. where I investigated how computational models of the I really believe you can forge your own path and be heart could be incorporated in the FDA drug approval surprised at where it takes you.

16 Inspired Research: News from the Department of Computer Science, University of Oxford Detecting extremist content online Mariam Nouh discusses new research to automatically detect radical content in social media.

In recent years we have seen several from the ISIS propaganda corpus. To order to understand the personality, examples of how terrorist groups create the radical language model, emotions, and the different and violent extremists exploit the we use different methods to extract psychological properties conveyed to internet to advance their agendas and the radical language from the corpus, readers in these articles. promote their actions. For instance, such as Term-Frequency Inverse- the Islamic State in Iraq and Syria Document-Frequency, and neural Furthermore, we make use of (ISIS) creating ISIS-branded media network model trained on word the radical language model and outlets and heavily utilising the embeddings to capture semantic. psychological model to detect pro- internet and popular social media ISIS tweets that are influenced by platforms to broadcast propaganda Moreover, inspired by research in their ideology. Unlike previous efforts, hoping to spread fear and radicalise fields such as linguistics, social our approach does not only focus individuals. Similarly, the recent attack science, and psychology that on lexical keyword analysis of the on a mosque in Christchurch (New suggest that the use of language tweets, but also adds a contextual Zealand) was broadcasted live on and the word choices we make and psychological dimension. Using multiple online platforms, and many in our daily communication, can machine learning classification of the attack plans were advertised act as a powerful signal to detect models, we are able to distinguish online a few days before the attack our emotional and psychological between radical and non-radical took place. Such incidents show the states, we extract psychological tweets. We validated our approach importance of identifying radical and properties from the radical corpus in in different experiments and the extremist content online in a timely order to understand the personality, results show that this method is manner. emotions, and the different robust across multiple datasets. psychological properties conveyed Among the most important feature In this research we aim to in these articles. Such psychological for distinguishing radical tweets is the automatically identify signals to properties include the ‘OCEAN’ or distance between the psychological detect online radical content. We ‘Big Five’ personality traits which profile calculated from the mainly focus on the ISIS group and are openness, conscientiousness, propaganda magazine corpus and we analyse the propaganda material extraversion, agreeableness, and the Twitter user. In addition, the us/ published in their online magazine, neuroticism. A number of studies them dichotomy which looks at the Dabiq. Dabiq is published by ISIS looked at the motivating factors total number of pronouns used with the purpose of recruiting people surrounding terrorism, radicalisation, (I, you, we, they) was significant. and promoting their propaganda and and recruitment tactics, and found ideology. Using this data source, that terrorist groups tend to target These identified signals are indeed we investigate what topics, textual vulnerable individuals who have critical to help improve existing properties, and linguistic cues exist feelings of desperation and displaced efforts to detect the spread of in these magazines. Our intuition is aggression. In particular, in research online violent extremist content and that utilising these linguistic cues into the recruiting tactics of ISIS radicalisation behaviour. This system extracted from extremist propaganda groups, it was found that they focus can aid law enforcement and Online should allow us to detect supporters on harnessing the individual’s need Social Network companies to better of ISIS who are influenced by such for significance. They seek out address such threats and help solve propaganda. vulnerable people and provide them a challenging real-world problem. with constant attention. Similarly, Finally, it is important to note here An overview of our two-phased these groups create a dichotomy and that such methods should always be approach is presented in Phase 1: promote the mentality of dividing the guided by a human analyst to oversee Radical Properties Extraction. First, world into ‘us’ versus ‘them’. In our the results of any automated models. we build a radical language model research, we extract psychological The complete research paper is and a psychological profile model properties from the radical corpus in available here: https://bit.ly/2D1QRAY

www.cs.ox.ac.uk Issue 15 Winter 2019 17 New standards to protect children online

A requirement of the Data Protection likely to be used by children. The ICO Our response welcomed the Act 2018 (which supports and produced a draft Age Appropriate recognition of responsible data supplements the implementation Design Code in May 20192. This draft handling as a critical issue in the of the EU General Data Protection provides practical guidance on 16 contemporary digital economy and Regulation) is the introduction of an standards of age appropriate design. as an issue that requires regulatory Age Appropriate Design Code. This The standards include: attention. We also particularly will set out the design standards welcomed the focus on children due required of online services and apps • Primary consideration given to their need for specific protections likely to be used by children in relation to the best interests of the child online. Promoting age appropriate to how these services and apps design in online services is extremely collect and process children’s data. • Consideration of the age complex; there are nuances around The background to this requirement range of users and the needs determining and applying relevant is the ubiquitous collection of of children of different ages age ranges for children, and it can be information about people whenever very difficult to identify and assess all they use online tools. This information • Transparent privacy potential risks and harms. forms ‘digital footprints’ and may information that is easy for contain revealing data about our users to understand However, in our view, the 16 draft identities, behaviours and so on. standards in the Design Code are • Avoidance of the use of comprehensive and well-defined. There are growing concerns that this children’s data in ways that is Several of them can be achieved collection of information can threaten detrimental to their well being with little practical complexity – for our privacy, data rights, and even our instance, the default for high privacy online safety. Children are spending • High privacy settings by and geolocation off. Based on the more and more time online; evidence default findings of our own projects, our suggests1 that in 2017, over half of response to the draft Code argued 3-4 year olds were online and nearly • Geolocation switched off by that fully serving the best interests all 12-15 year olds were. default of children requires developers to engage with young people, parents, Despite this, online services and apps • Avoidance of data sharing carers and educators. This will help are typically designed with the adult unless there is a compelling them to best understand children’s internet user in mind and there is a reason to do so. online experiences and form part of clear need for specific guidance for a community that serves to care and organisations offering online tools On publication, the draft Code was take responsibility for young people used by children so that they can opened up for public consultation online. better develop websites, platforms so that Individuals and groups could and apps etc that protect and uphold provide feedback on it. The work We also argued that developers and the rights on young internet users. done by members of the Human organisations should incorporate a Centred Computing theme at the long-term view of what constitutes The Information Commissioner’s University of Oxford takes a very detrimental use of data. Protections Office (ICO) has been tasked by strong focus on the well-being of should be put in place even when the UK government to produce this online users and several of our the harm or risk of harm cannot Age Appropriate Design Code. It projects have investigated children’s be discerned in the short term will ultimately recommend design use of technology. We were therefore as the long term and cumulative standards for data processing that very keen to submit a response to consequences of profiling can be providers should follow when they the draft Code. substantial. produce online services and apps

1 https://bit.ly/2r82XWM 2 https://bit.ly/2P6rBPH

18 Inspired Research: News from the Department of Computer Science, University of Oxford Engaging with your children: recipes for bringing up the ‘AI natives’, a brief background of the KOALA project

The rapid application of Artificial We first seek to understand the apps (which we collected from Intelligence technologies is children’s existing use of digital various information sources); but having a deep impact on our technologies and what they may expressed that they need more daily life, such as more accurate enjoy and struggle with. We are support to establish trust on such real-time traffic notifications or particularly interested to find out information or to make use of it. better support for our healthcare. how children may describe any risks However, AI not only makes our or barriers that they experience, This is positive feedback to an lives more efficient but may also so that we can communicate initial prototype which tries to influence the information we with children in a language that disrupt the existing market-leading consume online. AI is becoming far is most familiar to them [1]. We approaches, which mainly focus more ingrained in our lives than we have identified that children have on enabling parental control and could have imagined. an amazing ability to use a wide monitoring. KOALA Hero’s next step range of language to capture therefore will focus on providing At the heart of this new form their experience online (such as better support for parent-child of AI-powered digital life there ‘recommenders’, ‘youtubers’ or engagement by bringing in existing are our children, the so-called ‘autoplay’). parenting theories and providing ‘AI natives’, who grow up with better mechanisms for parents smart speakers from infancy and However, we also identified their and children to contribute and watch more YouTube than TV. We lack of ability to recognise implicit share their own experiences with wonder, how does this generation persuasive design features (such as technologies with their peers. We perceive and manage their life the game/video promotions) in the hope that in the future parents will with these technologies? How do technologies they interact with. This become empowered to support technologies impact their online lead to them spending more time each other. wellbeing? To what extent can our in front of screens or consuming children recognise and manage content less appropriate to their In the meanwhile, we would risks that they may encounter when age. encourage parents to: interacting with technologies? These are the fundamental research These are critical understandings • Engage with their children, by questions that Oxford Human that provided key inputs to the discussing what they enjoy during Centred Computing researchers set development of KOALA Hero their interactions with technologies. out to explore in the KOALA project (https://koala.web.ox.ac.uk/toolkits), • Start the conversation from a (http://koala.web.ox.ac.uk). an Android mobile app prototype young age, to develop a habit of co- that is aimed at supporting parents learning and co-experiencing a new The general goal of Kids Online in engaging with their young form of digital childhood Anonymity & Lifelong Autonomy children to discuss the use of • Pay more attention to what (KOALA) is to investigate the digital technologies and to co-learn children do in front of the screens, impact of personal data collection implications associated with the use instead of how long. practices used by mobile apps, of these technologies. upon the general well- being of More information about the KOALA young children. In early 2019, we presented our project: http://koala.web.ox.ac.uk. early app prototype to six local The project started in 2017, and we families. Our parent participants [1] Zhao et al. ‘I make up a silly have worked with over 40 families told us that they found existing the name’: Understanding children’s from 16 different local primary control mechanisms on the market perception of privacy risks online. schools. We focus on children aged insufficient for their needs, and they CHI 2019 between 6 and 10 because their welcomed KOALA Hero’s alternative experience with digital technologies approach to encourage parent-child has not yet been extensively engagement and co-learning; they studied in existing literature. We agreed that children must ‘learn’ are curious to learn about how this instead of simply being shielded age group may perceive the role of from risks. Parents also appreciated digital technologies in their lives. the additional expert reviews about

www.cs.ox.ac.uk Issue 15 Winter 2019 19 New research creates computer simulations of heart attacks

Hector Martinez-Navarro, a DPhil student in Computer Science, has essentially simulated the electrical consequences of a heart attack, as part of his research in the Oxford Computational Cardiovascular Science group, supervised by Ana Mincholé, Alfonso Bueno-Orovio and Professor Blanca Rodriguez. He has developed a computational pipeline for modelling and simulation of how the electrical activity of the human heart is affected by the block of an artery during a heart attack. The study has been published in Nature’s Scientific Reports.

Computational modelling and simulation are becoming ubiquitous in medical research. Widely exploited in many areas of engineering (eg aeronautics), computational modelling and simulation based on experimental and clinical data represents an alternative to animal and human experimenting, overcoming multiple ethical and technical limitations. Computational modelling and simulation are now also generating digital evidence for the development and evaluation of medical therapies. Both the European Medicine Agency and US Food and Drug Administration are currently developing new regulatory pathways that include modelling and simulation for both devices and drugs.

In this study, detailed mathematical models of human cardiac cells were incorporated into a whole-heart model, allowing the simulation of lethal arrhythmias and clinical signals such as the electrocardiogram. This approach was adopted by the authors to conduct highly biophysically- Changes in detailed simulations of the human heart electrophysiology under ischemic the simulated conditions, when cells in the region affected by coronary occlusion are electrocardiogram in deprived of oxygen and nutrients. These are computationally expensive, the virtual human heart requiring the use of high performance computing, such as ARCHER (http:// affected by coronary archer.ac.uk/), the UK’s largest supercomputing facility. occlusion. Transmural ischemia (when the whole thickness of the heart wall is affected) [left] causes larger electrocardiographic abnormalities than subendocardial ischemia (which is when the inner layer of the heart muscle During a heart attack, a coronary artery is occluded or obstructed, and blood is affected) [right]. In irrigation of the cardiac muscle is compromised. This leads to ischemia in both cases, the dashed the affected tissue: the tissue is starved of oxygen and can eventually die. line corresponds to a Myocardial ischemia leads to abnormalities in the electrical propagation in control non-ischemic the heart muscle, potentially causing life-threatening arrhythmias (irregular simulation. LAD heartbeat). Variability in symptoms and arrhythmic risk caused by ischemia occlusion: occlusion remains a challenge in the diagnosis and management of patients. Some of the Left Anterior ischemic patients exhibit ECG abnormalities, whereas others show no clear Descending artery, the changes in the ECG. The relationship with arrhythmic risk is unknown. most important artery of the human heart in blood supply. continued on next page

20 Inspired Research: News from the Department of Computer Science, University of Oxford from previous page

Arrhythmia around the region of ischemia (figure- of-eight pattern) in transmural ischemia, affecting the whole ventricular wall from the ventricular cavity to the epicardial In this study, the authors create multiple virtual scenarios of ischemia surface. reproducing the variability observed clinically, using experimental and clinical data for model construction and validation. The scenarios considered included different sizes of ischemic region, extensions across the ventricular wall, and locations through consideration of the occlusion of different coronary arteries. Fully transmural ischemia scenarios (ischemia affecting the entire myocardial wall) reported remarkable changes in the simulated electrocardiogram, and typical arrhythmic patterns around the afflicted region.

Subendocardial ischemia, on the contrary, only affects the inner part of the myocardial wall, close to the ventricular cavities (eg the endocardium). Even though this only produced mild or negligible changes in the simulated electrocardiogram, the occurrence of arrhythmias was still very high, in agreement with clinical evidence. Through detailed analysis, arrhythmias in subendocardial ischemia are shown in this study to be linked with a different mechanism, as they are sustained transmurally within the myocardial wall, rather than around the region affected by the coronary occlusion.

Transmural arrhythmia in subendocardial ischemia. The top panel shows the cardiac surface, providing no information about the arrhythmic pattern. The bottom panel shows the transmural pattern with a detailed view, where the arrhythmia develops in the form of a spiral waves inside the ventricular wall.

Thereby, the simulations show that dangerous transmural patterns of arrhythmia can occur in patients with only subtle changes in the electrocardiogram. This would explain the still high arrhythmic risk reported in patients suffering from silent ischemia episodes, typically produced by incomplete occlusion of their coronary arteries.

The computer modelling and simulation technology and results in this study are an illustrative example of the successful application of in silico models in biomedical research. The human heart models and simulations in disease can be used as a test bed for therapies in in silico trials, to inform further experimental and clinical studies.

Read more: https://bit.ly/2qjIBts

www.cs.ox.ac.uk Issue 15 Winter 2019 21 Virtual Assay: user-friendly software for in silico drug testing By Elisa Passini

In the last few years one of development thus leading to better have demonstrated the ability of the research lines within our and safer medicine. Virtual Assay is the software to predict the risk that Computational Cardiovascular also easy to use by non-experts in drugs would cause abnormal heart Science Group has been focusing modelling and simulations providing rhythms in patients with almost 90% on how computer models can be a very efficient simulation engine (1 accuracy, whereas similar studies used to predict drug safety and drug trial takes about 5 minutes for conducted in animals showed 75- efficacy for the human heart. Driven 100 cells, using a modern PC). 85% accuracy. by the idea that drug testing could be performed by a computer rather The first version of the Software The potential of the software for the than on animals, our researchers was released in 2014, and after replacement of animal experiments developed the Virtual Assay almost 5 years of iterations, guided has been recognised by two Software which provides a user- by the feedback received from international 3Rs prizes, awarded friendly graphical user interface to our pharma collaborators testing to Dr Oliver Britton (2014) and run human in silico (computational) the software, Virtual Assay V.3.0 Dr Elisa Passini (2017). Through drug trials in populations of heart is now ready. A free academic dissemination of our results, we models. Predictions obtained licence and a commercial licence have also contributed to changing with the software will guide drug are available through Oxford the culture in pharmaceutical discovery in the future, identifying University Innovation. Multiple industries, opening their minds to potential side effects for the human evaluation studies have been computational approaches. heart in the early stages of drug performed using Virtual Assay, and Read more: https://bit.ly/2Lq0yhf Beyond Academia: what we have learned from our experience with industry By Elisa Passini Since cardiovascular side effects a very different direction compared aspects of pharma, and also the are still one of the major problems to what was planned originally, and opportunity of appreciating more during drug development, our where last-minute deadlines happen applied and structured style of research on human in silico drug trials all the time. Also, industry tends to research. It also gave me visibility is very relevant to pharmaceutical take into account a larger time scale and personal recognition, through companies. By disseminating when evaluating new methodologies. the International 3Rs prize and the our results in industry-oriented They want to make sure that what is Technological Innovation Award conferences, we have been able to working today, will also be working in (Safety Pharmacology Society). This attract the attention of the people a few years, and sustainability in the also led to numerous invitations working in the safety pharmacology long term is crucial. to present my research to national departments, and we have and international meetings, to established successful research Third, changes take time. The different audiences with varied collaborations with more than seven CiPA initiative (https://cipaproject. expertise. Managing and nurturing big pharma companies in the last org), sponsored by the US Food the industrial collaborations, as well six years. This has given us the and Drug Administration and other as contributing to large industry-led opportunity to appreciate many organisations, started in 2013, with projects/consortiums, gave me the differences between academia and the aim of implementing a new opportunity of establishing my role industry. paradigm for drug safety assessment, within the group, maturing into a also including computer models. more senior research position. All First, every new methodology is The proposed implementation this will have implications for future evaluated in terms of quality, time and was supposed to happen by 2015 grant/fellowship applications, where cost, and strong evidence must be and now, 4 years after this – very I will be able to bring on board provided to demonstrate its advantages optimistic – deadline, things are still industrial collaborators, providing compared to the existing ones. work-in-progress. However, changes support and experimental data, are slowly happening, and we can which are essential to develop our Second, a good management of proudly say we are making our computational models. time and deadlines is essential. Even contribution. before starting a job, we are required Finally, from a personal point of view, to provide an accurate estimate of For me – as an early career it is very rewarding to see that our how much time/work will be needed researcher – establishing good research can achieve impact beyond to complete it, and any publication – collaborations with industry has academia, pioneering changes that even a conference abstract – needs been very important. It made a big could lead to a revolution in the way to go through an internal review difference in terms of extending my drug safety and efficacy testing process, which could take up to a network of contacts, opening the are currently performed by pharma month, before submission. Things way for future research projects industry, with concrete implications are quite different in academia, and grant applications. It gave such as the reduction of animal where a project might end up taking me insights on the operational experiments.

22 Inspired Research: News from the Department of Computer Science, University of Oxford Training autonomous vehicles using real-life human behaviour

By Stuart Gillespie

Driverless cars are on their way – there’s little doubt about that. But before they hit the UK’s roads, they need to be tested in realistic simulations to ensure that this transformative technology will be a safe and positive addition to our lives.

Latent Logic, an Oxford University with imitation learning. Shimon ‘Autonomous vehicles are improving spinout company, is helping enable says: ‘Our models extract the “latent rapidly, but we are still some this vital testing with technology that logic” behind real-life examples way from realising the dream. teaches autonomous systems using of natural human behaviour. As a It is not just about perfecting real-life examples of natural human result, they can respond realistically existing technology – there remain behaviour on the roads. even in new situations. fundamental unsolved problems in building sufficiently robust Professor Shimon Whiteson, co- ‘We use computer vision to collect autonomous systems. At Latent founder and chief scientist at Latent these examples from video data Logic, we believe our technology Logic, explains: provided by traffic cameras and can play a critical role in addressing drone footage. We can detect road those unsolved problems. ‘Autonomous vehicles must be users, track their motion, and infer tested in simulation before they their three- dimensional position ‘In addition, our technology has can be deployed on real roads. To in the real world. Then, we learn to numerous other applications. make these simulations realistic, generate realistic trajectories that Situations in which you want it’s not enough to simulate the road imitate this real-life behaviour.’ socially normative behaviour are environment; we need to simulate great candidates for learning from the other road users too: the human By providing a service that enables demonstration, as is robotics – from drivers, cyclists and pedestrians with better training and testing of factories to warehouses to homes. which an autonomous vehicle may autonomous vehicles, Latent You can also think about video interact. Logic hopes it can hasten the games, where people might want to safe introduction of what will play against bots that can imitate the ‘Latent Logic is using a machine be life-changing technology. style of their favourite professional learning technique called imitation Shimon adds: ‘While many of the gamers. The sky’s the limit.’ learning to build realistic human biggest players in this market are behaviour models. These models international, there is also a lot Latent Logic grew out of a research make it possible to test autonomous of energy in the UK in this sector, project which trained semi- vehicles quickly and robustly.’ and Oxford is a hotbed of talent autonomous telepresence robots to and entrepreneurship in machine behave in a socially ‘normal’ way. Latent Logic grew out of an EU- learning, robotics and autonomous funded research project that trained vehicles. Read more: www.latentlogic.com semi-autonomous telepresence robots to behave in a socially ‘normal’ way. Since it’s difficult to quantify what is meant by socially normal, it’s much easier to train such systems to imitate the behaviour of humans acting in a socially normal way. Shimon decided to explore the commercial potential of this technology, recruiting postdoctoral researcher Dr João Messias to be co-founder and chief technology officer. Kirsty Lloyd-Jukes then joined the company as CEO.

The technology works by combining state-of-the-art computer vision

www.cs.ox.ac.uk Issue 15 Winter 2019 23 The knob attack on Bluetooth

By Professor Kasper Rasmussen

Bluetooth is a wireless communication technology that is present on most modern devices. Every mobile phone has a Bluetooth chip and Bluetooth is used by people around the world every day, eg, to enable your phone to communicate with a headset, your car, or your smart watch. It is the protocol of choice for connecting devices together and is used in everything from kitchen appliances to implantable medical devices. If you buy a new Tesla and want to open your car with your phone (yes that is a new feature Tesla offers), your car and phone communicate via Bluetooth. Anytime two devices are meant to communicate wirelessly over a short distance Bluetooth is the obvious choice because of its ubiquity and the fact that it offers a secure connection through which to communicate. is started a new session-key is value can be anything between 1 Bluetooth offers several different generated in order to encrypt and and 16 bytes. The problem is that modes of communication, including authenticate the data exchanged this negotiation is not protected those that offer a confidential during the session. It is in this in any way, and it therefore allows and authenticated link. When session-key establishment an external adversary who is two Bluetooth devices want to procedure that the Bluetooth present during the negotiation, to communicate securely they first standard has a critical flaw. When force the devices to generate a have to be ‘paired’. Pairing is a two devices wish to establish a session-key with only a single byte procedure by which the two devices new session-key, they do so by first of entropy. That means that the agree on a long term secret key that generating a preliminary session-key resulting session-key only has 256 forms the basis for future secure utilising freshly generated random possibilities, all of which can be communication. This key is often numbers as well as the Bluetooth quickly checked by the adversary authenticated through an out-of- address, and the long term pairing and thus it allows the adversary to band channel, for example by a key. These parameters are passed obtain the session-key. user reading digits off the screen of through a key derivation function to one device and typing them into the generate the preliminary session- Once the session-key has been other device to verify that the two key. This preliminary session-key obtained an adversary can choose devices do in fact share the same is a perfectly good 128 bit long to passively listen in on the session, key. session-key that could be used ie, eavesdrop on the data being to secure the session adequately. shared, or the adversary can choose Using this long term key, two However as part of the session- to participate in the conversation, devices that wish to communicate, key generation mechanism there pretending to be either one of the can now establish a new session. is an additional step that requires endpoints. The adversary listen in, A new session is established the devices to reduce the entropy and inject arbitrary data into the every time two devices wish to of the freshly generated session- connection, for the duration of the communicate after having been key to a value that the devices session. disconnected. When a new session must negotiate ahead of time. This continued on next page

24 Inspired Research: News from the Department of Computer Science, University of Oxford from previous page the attacker can simply jam the In response to our findings the ongoing transmission and wait for Bluetooth working group has This attack is not against a one of the devices to re-establish released an errata to the standard particular implementation of the the connection. In re-establishing and will no doubt make sure that Bluetooth standard but rather the connection the devices will start future revisions of the standard against the standard itself. For this a new session, and the adversary either removes the entropy reason this vulnerability is present can interfere with the session negotiation all together or mandates on every single Bluetooth enabled establishment and make the devices that devices choose the maximum device in the world, which means derive a weak key, that in turn can value of 16 bytes (128 bits). The every single mobile phone, every be immediately cracked. International Consortium for single laptop, every single smart Advancement of Cybersecurity on watch, and most cars produced An attacker can basically force the Internet (ICASI) — an industry in the last 15 years. Millions and victim devices to start a new led coordination body founded millions devices are affected. session whenever he wants, by Intel, Microsoft, Cisco, Juniper To make things worse, it is far however he still has to be within and IBM — has spent the last from trivial to roll out a fix for the Bluetooth range to make the attack eight months trying to develop a problem. The Bluetooth standard work. This range is often listed workaround for the various devices mandates that the protocol is as around ten metres, and while and operating systems that these handled not by the operating that would expose a connection companies produce. That means system of the device, but by a to attack from anyone in the same that while the attack itself is not separate Bluetooth chip-set which cafe, office or apartment complex, prevented, the operating system hides the session establishment it does limit the attack surface. can try to detect if the Bluetooth details from the rest of the system. Or it would, if the ten meter range connection was subject to such This means that the only way to was really true. The Bluetooth an attack, and if so, tear down the change the behaviour of a Bluetooth range is ten metres only when connection and try to reconnect. device, is to upgrade the firmware using a normal antenna and normal We are currently working on an of the Bluetooth chip-set, which equipment, however with a very analysis of the effectiveness of this is not always possible, and even if simple directional antenna, that workaround. a device allows this, not all users anyone can construct from a tin can would be willing or able to perform and a couple of wires, the range of For the foreseeable future, it would such an update on all their devices. Bluetooth can be increased up to be wise to treat any Bluetooth Even though this attack was a kilometre or more without much connection as insecure, thus if you disclosed to the Bluetooth working effort. This of course assumes that need a secure connection you must group, and they had eight months the adversary is able to point their provide your own security on the to come up with a workaround, directional antenna accurately but application layer. many devices will continue to be it certainly increases the potential vulnerable for many years to come. range of the attack.

In order to take advantage of this vulnerability, an adversary must be within Bluetooth range when two devices are starting a new session. This seems like it would limit the attack surface somewhat, and it does, but not as much at it might first appear. First of all, starting a new session is something that happens all the time, and if an attacker arrives late, say he arrives in a cafe and there is already a Bluetooth connection in progress,

www.cs.ox.ac.uk Issue 15 Winter 2019 25 Bayesian deep learning for all humankind

By James Parr, Director - Frontier Development Lab (FDL) Europe Hosted at Kellogg, Oxford

AI research is exploding. So too are the kind of problems it is inherently good at tackling; surfacing insight from large or noisy datasets; spotting anomalies; fusing data and reducing the time and complexity of making informed decisions.

Deep learning/machine learning has emerged on the Bradley Gram-Hansen, Lewis Smith, Tim Rudner, world stage at just the moment the world needed it the Tim Seabrook, Richard Strange and many other most. However, the journey to deployed and trusted AI talented researchers from around the world who have workflows is only just beginning. pushed the envelope of machine learning application.

The barriers are often systemic (such as the perennial For all Humankind challenge of AI ready data, compute costs and access to talent) but some barriers are also about the fundamental During FDL, the teams work intensely over a period of nature of neural nets. Can we trust them? Can we explain eight weeks, supported by world-class mentors and how they’ve reached their conclusions? Can we quantify industry partners as well as the space agencies - with their uncertainty? remarkable results. A particular focus has been climate and disaster response, where earth observation data is In relations to the questions we are seeking to explore, analysed by machine learning to give improved situational the Frontier Development Lab (an applied AI research lab awareness - in other words, more timely information to working in partnership with European Space Agency and humanitarian agencies, governments and communities in Oxford in Europe; and NASA in the US, alongside leaders case of natural disasters. in commercial AI, such as Google Cloud and Intel) was very fortunate to meet Yarin Gal: an Associate Professor Rapid damage assessment remains a problem in disaster of Machine Learning at Oxford’s Department of Computer response. FDL teams have used machine learning Science. Yarins particular specialty is at the intersection techniques on Earth observation data to make substantive of neural nets and probability theory - Bayesian Deep improvements on inundation prediction (time to flood) Learning - key to making AI a useful applied tool. and prototype the use of machine learning on board a spacecraft in orbit to map a flood extent, enabling maps Eight years after the ‘Deep Learning Big Bang’ in 2012, to be computed in space and sent to first responders when neural nets running on GPUs where shown to match on the ground. They have used AI’s ability to fuse data human capabilities in image recognition for the first time, to assess building damage, but before all that, you need the field is starting to branch off into specialty areas such to know where the people are. It is hard to believe that as reinforcement learning and adversarial techniques. In in 2019 we still don’t have access to detailed maps of response to the relentless pace of innovation, AI labs have vulnerable communities. started to specialise. Yarin and the Oxford team, helped steer FDL’s specific niche into creating deep learning/ Using machine learning to create a census of machine learning solutions that could be explained - a Informal Settlements crucial competitive edge when creating solutions for the exacting expectations of a wary space industry: Is it safe? One-third of the world’s urban population lives in informal Can we interpret what’s happening? Can we be confident settlements; these are the most socially and economically in the outcome? vulnerable communities on Earth. Sometimes they are located within city limits. Sometimes in very rural areas. FDL was established to apply cutting edge machine People living in these areas often have no security of learning for the good of all Humankind. However, it is the tenure, they often lack basic services - such as medicine 300 year old Bayes Theorem which is the key to making or immunisation, or infrastructure, such as waste disposal this dream viable. or clean water. Housing rarely complies with building safely standards. Moreover, due to zoning issues and FDL’s success has been due to the hard work of Oxford Researchers such as Atılım Güneş Baydin, Steve Reece, continued on next page

26 Inspired Research: News from the Department of Computer Science, University of Oxford from previous page with medium resolution, multi-spectral optical and radar satellite imagery (SAR) to automate the creation of earth other oversights, often these settlements are simply observation-based disaster impact maps (Fig 1) for first overlooked - becoming hotspots for poverty and disease. responders, affected communities and aid/assistance coordinator. FDL used machine learning to combine spectral information from satellites and ground truth information This machine learning (called Multi3Net) fuses optical on building materials to automate mapping of informal and radar imagery to penetrate clouds and identify settlements on an unprecedented scale and frequency changes on the ground. The method consists of multiple (referred to as temporal resolution). streams of encoder-decoder architectures that extract The result of this approach is a tool that can provide the damage from medium-resolution optical images, census information on a country, or even continental radar data and spatial information from high-resolution scale at virtually no cost, with a very high degree of images. The resulting representations are combined on a temporal frequency (every 5 days). This work will hopefully single map that accurately captures damage assessment contribute to projects such as WorldPop (https://www. on the ground. worldpop.org). These capacities are not only crucial for aid agencies such as UNICEF and the WHO, but also the Red Multi3Net exceeds state-of-the-art methods for Cross and the Disaster Emergency Committee (DEC), who assessing flood, earthquake or hurricane damaged often have poorly defined maps of disaster prone areas. structures, moreover this can be performed using freely available medium-resolution data instead of significantly Using machine learning to improve disaster response more detailed (and expensive) very high-resolution data. This approach - in theory - enables damage assessment During a hurricane, earthquake or flooding event, the to happen during the disaster itself, helping governments experience on the ground can be compared to the ‘fog’ to better coordinate medium and long-term financial described in battle zones - where a combination of stress, assistance programs for affected areas. The bottleneck breakdown in trusted information services, uncertainty remains the frequency of satellites providing global and panic, conspire to exacerbate the immediate coverage, but plans are in pace for mega constellations challenge of the disaster itself. Moreover, despite our of satellites that will allow the ‘just-in-time’ insight century’s rapid acquisition of peta-scale geospatial data discussed here. - even in 2019, the translation of this data into insight on the ground during a disaster is still woefully slow. Often The future: A Mission Control for Planet Earth it is only after the Disaster has past do first responders really get an understanding of the damage to property, The concept of a ‘Mission Control’ for Earth emerges infrastructure and sadly, human life. Lack of information as an aspirational endgame for these technologies, with about affected communities and the level of damage also ML enabling data fusion, classification, prediction and hinders efficient response coordination (such as funding) change detection. This vision of a joined-up ‘planetary by authorities and insurance companies. So the question stewardship’ still requires a number of innovations in is can machine learning help? workflow to optimise AI techniques, before the human race sees the benefits. FDL, with the help of Oxford The FDL team was able to show the potential of data Computer Science, is already well on its journey to making fusion to enrich high-resolution optical satellite images this happen.

Fig 1: A future Disaster Response Dispatcher based on Fig 2 AI / Earth observation approaches(Source: FDL)

www.cs.ox.ac.uk Issue 15 Winter 2019 27 About Time: Changing security behaviour through smartwatch games

By Meredydd Williams

Smartwatches have grown in popularity over the past indicated their final behaviour. By comparing the pre- decade. Whether Apple, Android or Fitbit, millions test and post-test results, we explored our game’s of users are strapping functionality to their wrists. potential. In the treatment group, individuals increased Indeed, smartwatch sales have increased by more than their usage of passwords and permissions. Control- 60% in 2019. These devices provide connectivity and group behaviour failed to change, and remained convenience in a mobile form factor. However, as with unprotective. This suggested an opportunity for most digital technologies, these benefits come with a smartwatch security games. risk. While these results were encouraging, self-reporting is To provide such functionality, smartwatches access a prone to bias. Therefore, we chose to explore behaviour range of data. Some of this, such as texts and contacts, change empirically. To achieve this, we implemented may come from a synched phone. Other apps might our app on an Android smartwatch. This game also be installed directly on the device. As watches are included thieves and security challenges. However, strapped to the wrist, they can also collect GPS and these tasks were customised based on each user’s heart-rate data. Some models do include security apps. Gameplay also updated in response to real-life settings. However, despite the risk, users tend to avoid behaviour, providing a positive feedback loop. this protection. After all, what threat could my watch pose? The challenge is persuading individuals to guard For a comprehensive evaluation, we conducted an their devices. eight-week longitudinal study. At the start, we gave 10 students an Android smartwatch. In the 18-day pretest Awareness campaigns tend to be the default response. period, we recorded their baseline use of settings. This Cyber Streetwise adorned tube stations and TV screens was logged through an ethically-approved tracking across the country. However, despite a million-pound app. In the gameplay phase, five treatment participants investment, the campaigns had little effect. The reason received the security game. To mitigate bias, the control is clear: awareness is not enough. Individuals also group were given a similar app with generic challenges. require education and incentivisation. Without the Gameplay was followed by an 18-day post-test period, former, they lack the protective skills. Without the latter, enabling analysis of behaviour change. they will be apathetic to act. Serious games take an active approach, embedding incentives through positive Prior to gameplay, all participants failed to protect reinforcement. Meta-analyses suggest that apps have their data. Indeed, no users disabled GPS tracking greater efficacy than awareness campaigns. Therefore, or restricted their permissions. However, the groups we developed the first game for smartwatch security. greatly differed after gameplay. Treatment participants increased password usage and decreased GPS usage. Our app was designed through best practice from This suggests that games can indeed encourage learning science and behavioural psychology. However, smartwatch protection. In contrast, control group due to the scarcity of smartwatch games, we were behaviour failed to change. Throughout the eight-week in uncharted territory. For this reason, we began by period, 80% of control participants did not restrict a developing an online prototype. Our game challenged single setting. Therefore, it appears that users will take users to perform tasks on a simulated smartwatch little action unless informed and incentivised. interface. Players navigated their way through several levels on a map. En route, they encountered thief Awareness is not enough; individuals must receive characters who set them security challenges. If users education and encouragement. We suggest the use of updated their settings correctly, they won extra points. serious games as a complement to security awareness This design highlighted issues, enabled practice and campaigns. Read more: https://bit.ly/2sgeXpx incentivised protection.

To assess and refine the prototype, we recruited 504 smartwatch owners. At study commencement, all users indicated their degree of settings protection. One half, our treatment group, played the game one week later. The control half did not, as to mitigate confounding factors. Another week after gameplay, all participants

28 Inspired Research: News from the Department of Computer Science, University of Oxford