Human- Computer Interaction at Cambridge

Document Sample
Human- Computer Interaction at Cambridge Powered By Docstoc
					Natural Computing Applications Forum

                                       Interaction at
                                       3 – 4 September 2003

      Many interaction                         uman-Computer Interaction (HCI) will be          Interfacing. Prof. Chris. Bishop (Microsoft Research)
                                               the theme of this year’s autumn meeting,         will be giving ‘An Introduction to Mixture Models,
          modalities with                      which will be held in Cambridge on 3 and 4       EM, and all that’ to begin the second day of the
                                       September. This is the first joint meeting of NCAF       meeting. Both are well known in their respective
      computers will be                and Microsoft Research, which is also generously         fields and are excellent speakers, promising an
                                       sponsoring the event. We have defined HCI in the         exciting start on both days.
     discussed, ranging                widest sense, covering not only technical aspects            Other presentations will include reports on
                                       of humans interacting with machines, but also            progress of the VIBES project (Variational Inference
   from user interfaces                philosophical, design and psychological aspects of       Engine for Bayesian Networks) by J. Winn (Microsoft
                                       HCI.                                                     Research), job-shop scheduling using software
                for digital                 Many interaction modalities with computers will     agents by P. Swann, (Rolls-Royce) and co-operative
   photography, hands                  be discussed, ranging from user interfaces for digital   behaviour of complex systems by Ian Nabney
                                       photography (K. Wood, Microsoft Research), hands         (University of Aston). In short, the autumn
        free interfaces to             free interfaces (D. MacKay, University of Cambridge),    programme promises to contain many exciting talks
                                       to Brain Computer interfacing (P. Sykacek, University    on a large variety of HCI tools by well-known experts
       Brain Computer                  of Oxford), computer vision (A. Criminisi, Microsoft     in their fields. If this is not enough we have many
                                       Research), and computer imaging (P. Perez,               presentations on non-HCI related issues which will
              interfacing,             Microsoft Research).                                     reach the parts that other talks could not reach.
                                            Each day will commence with a tutorial. On the
  computer vision, and                 first day, Prof. R. Spence (University of Cambridge),    Dr. Iead Rezek
    computer imaging.                  will cover the principles of Human Computer              University of Oxford

                                       Responding to the survey
                                           n order to return to some of the core elements of    attractive to students planning to utilise

                                       I   NCAF’s original remit and in response to returns
                                           from the recent Networks readers’ questionnaire,
                                       it is proposed that the Winter 2003/4 NCAF meeting
                                                                                                computational methods in their research.
                                                                                                    The provisional date of the meeting is 7–8
                                                                                                January 2004 to fit in with the academic calendar
                                       will be themed on Fundamentals of Natural Computing.     and the industry post-Christmas lull. The basic
 ■   Natural Computing is              It is hoped that during the first day there will be a    details of the meeting, including the majority of
     Safe in our Hands                 tutorial on the fundamental techniques used in           speakers, will (honest!) be available for members to
 ■   Puzzle Corner No. 24              machine learning and artificial intelligence             peruse at the September meeting and the finalised
 ■   Diary Dates                       applications. To consolidate this, a workshop on         agenda will be available by the end of that
 Back Page                             one or more of these techniques is planned for the       month…watch this space!
 ■ Researching complexity              second day. This is an ideal opportunity for people
     in clusters                       new to Natural Computing techniques to be                ’Bert Bullen
                                       introduced to the area. It may be particularly           BAE Systems

         Official Newsletter of the Natural Computing Applications Forum Edition 38 – July 2003
                            Natural Computing is
                            Safe in Our Hands
    …the highlight of             June 2003. Outside lane of the M5. Smile on my      High Integrity Software Engineering department at

 Graham’s day, as he        4     face. All the hassles of the world left behind
                                  me. Yes it’s time for another couple of days in
                            the invigorating world of the latest NCAF meeting. I
                                                                                      the University of York. Zeshan’s talk, entitled
                                                                                      ‘Developing Artificial Neural Networks for Safety
                                                                                      Critical Applications’, included a discussion of the
                            have to confess I always look forward to these            importance of addressing the safety lifecycle in
  transformed himself       events, with their mix of technology, ideas and           safety critical system development. He described a
                            people, all set against a relaxed, friendly               hybrid ANN design approach that had been
      into the mystery      background. (To see just how relaxed, see later           developed at York, which enabled some of the
                            under ‘Social Event…’).                                   advantages of neural networks to be delivered
       man of natural           The June event was held at Exeter University’s        within the strict constraints of safety critical
                            Crossmead Conference Centre, and Richard                  development.
computing, the Great        Everson of Exeter University had put together an              Zeshan was followed by Alastair Faulkner (CSE
                            appetising mix of papers and events. As usual with        Limited & University of Warwick), who raised the
       Enchilada, and       NCAF meetings, the event had a mix of themed and          issue of data integrity in safety related systems.
                            general papers. The theme for June’s meeting was          Alastair pointed out that when we are developing
   entertained us with      ‘Safety Critical Systems’, a domain offering rich         and managing systems, it is too easy to concentrate
                            pickings for the natural computing world. The use of      on the integrity of the design, verification and
      his latest trick of   neural networks in safety critical systems has often      validation of the software, and forget that many
                            been considered an anathema, and I was interested         applications are critically dependent on the data
digital prestidigitation    to see how these two often diametrically opposed          they use, and quality, development, maintenance
                            and incompatible, worlds could be reconciled.             and assurance of the data is often outside the
  and managed (after            The meeting opened with a talk on ‘Bayesian           control of the developer or user. He recommended
                            Reverse Jump Markov Chain Monte Carlo’ (RJMCMC)           that the data used in and by safety related systems
         an inordinate      Methods for Safety Critical Systems’ by Wojtek            should be subjected to similar hazard and risk
                            Krzanowski (University of Exeter). Wojtek outlined        analyses that we apply to the system design and
    amount of faffing,      work aimed at developing classifiers that could also      that attention should be paid to the structure and
                            deliver measures of the confidence of their               validation of data to ensure that potential errors are
                            classifications, as required by safety critical and       uncovered and resolved. In addition to an excellent
       assistance from      safety related applications. Building on datasets         talk, Alastair also managed the previously unknown
                            from aircraft tracking and trauma treatment domains,      achievement of finishing with a quote from Hiram K
        his accomplice,     Wojtek and his team had developed techniques              Hackenbacker, one of the best-known technical
                            based on RJMCMC to improve a KNN classifier, and          brains of the last century! (From Thunderbirds, for
     Guacamole, and         early results were encouraging. Further work was          those who had forgotten.)
                            under way, addressing enhancements to distance                After tea, two NCAF stalwarts Mandy Bradley
  complete suspension       measures and feature detection.                           and Chris Satchwell (Technical Forecasts Limited),
                                The next paper, by Jonathan Fieldsend,                presented a general paper on the subject of
         of disbelief by    (University of Exeter), addressed the problem of          ‘Forecasting Values of Commercial and Residential
                            ‘Tuning Legacy Critical Systems with Multi-objective      Property Using Non-Linear Mathematical and
      the audience) to      Optimisation’. Jonathan’s talk described work             Statistical Techniques’. As well as describing the
                            underway to tune an existing application, Short           approach that Technical Forecasts have been taking
      predict the value     Term Conflict Alert (STCA), which is in use by many       to forecasting, employing an array of multiple
                            national air traffic services to predict when two         models and data sources, Mandy and Chris
      of a hidden card      aircraft appear to be likely to get too close.            presented an important lesson on dealing with
                            Jonathan’s work uses multi-objective evolution            errors and uncertainty in source data.
          from a pack.      strategy to explore the boundary of locally optimal           A short break allowed me to slip in a quick
                            solutions, generating an achievable Receiver              advert for the work I am undertaking concerning the
                            Operating Characteristic (ROC) curve, and so              non-technical aspects of developing AI based
                            allowing an informed selection of operating point,        systems. (For more details see
                            trading off likelihood of true positives against false
                            positives to best suit the application. Initial work on       Next – the highlight of Graham’s day, as he
                            trial databases had proved encouraging, and the           transformed himself into the mystery man of
                            team are starting to address STCA optimisation.           natural computing, the Great Enchilada, and
                                After coffee Paulo Lisboa (Liverpool John Moores      entertained us with his latest trick of digital
                            University) presented a paper titled ‘Confidence in       prestidigitation and managed (after an inordinate
                            Neural Networks: Methodological Issues Arising from       amount of faffing, assistance from his accomplice,
                            a Review of Safety-Related Applications’. Paul’s talk     Guacamole, and complete suspension of disbelief
                            looked beyond the technical issues of developing          by the audience) to predict the value of a hidden
                            safety critical systems, and addressed all the related    card from a pack. Luckily, it ended before rigor
                            issues of making successful applications, including       mortise took hold.
                            good practice throughout the software development             That completed the formal part of the day – all
                            cycle.                                                    in all a great success – not too many equations and
                                               Safety Cycle                           a good focus on real practical issues. So it was off to
                                                                                      the social event – canoeing down the river to one of
                            After a splendid lunch, and plenty of chances to          Exeter’s best pubs, the Double Locks. But… the
                            chat, make new contacts and renew old friendships,        weather was against us, and instead of a sun kissed
                            we resumed with a talk from Zeshan Kurd from the          evening enjoyed against the sound of the quiet
splash of paddles in the calm water, we were faced      demonstration and a highlighting of the problems
with the prospect of rain and Richard had made the      of the sensitivity of the system to source/sensor       PUZZLE CORNER
sad but sensible decision of cancelling the canoes.     position rounded off an excellent talk.
Undaunted, we set out for the pub by taxi, only to          The last speaker of the morning, Ali Hessami        Number 24
find that the weather was chilly but dry all evening!   (Atkins Global), treated us to a fascinating and        Being pushed out of a
No problem! We had a great time, quaffing the           thought provoking presentation entitled ‘Ancient        helicopter hovering six feet
Double Locks’ excellent beer and wolfing down           Wisdom, Modern Solution’. Drawing on the ideas of       above a minefield in the
large portions of their home cooked food. Once we       Zarathustra, Heraclitus and Tao, Ali described          middle of a pitch-black night
were satiated we walked back into town to the           Weighted Factor Analysis (WeFA), a new approach         was not Lisa’s ideal way of
bowling alley, and started the second event of the      to problem identification, exploration and              ending the evening.
evening! Here we discovered that Mark ‘Bowling          resolution, based on a guided hierarchical              Nevertheless, that’s where she
Balls’ Cheeseman was a bit of an expert (casually       decomposition of a problem, and allowing all            found herself. A sandy strip a
tossing in that he used to bowl for England or          aspects to be considered. WeFA has been used in         mile wide and many miles
something like that). Mark wiped the floor with us in   anger on real world problems, and continues to be       long, Lisa knew it was the no-
the first game, but we were pleased to see that his     developed into a practical tool.                        mans-land between the
skills miraculously evaporated in the return match.         After lunch, the final two speakers were both       elephant grass of the Hellarwi
A short walk took us back to the conference halls of    from Oxford University. Natasha Chia covered the        pigmy tribe and the verdant
residence and to well earned sleep (except for a        topic of ‘Neuromorphic Engineering’, whose design       pastures of the bull-
diligent few who undertook the advanced card            principles and architecture are biologically inspired   worshipping Nagumba.
                                                                                                                   Out of a warped sense of
techniques school.)                                     (a concept introduced by Carver Mead). The
                                                                                                                fair play, the helicopter crew
                                                        ultimate goal is the realisation of future thinking
                                                                                                                had given her a compass, a
       Active Learning Algorithms                       machines, and the potent blend of neuroscientists,      pathetically weak flashlight
                                                        electronic engineers and mathematicians are             and a yardstick. Using these
Day 2 dawned all too early, but we recovered over a     hoping to deliver this through more closely             tools Lisa was capable of
slow and excellent breakfast, before starting the       modelling the spiking behaviour of real neurons.        accurately following any
serious business of the day. The first paper was        The meeting culminated in Rizwan Choudrey’s             desired path. Unfortunately
presented by Colin Campbell (Bristol University).       ‘Beyond Independent Component Analysis:                 she had no idea how the
Colin presented his work on active learning             Hierarchical Models’ talk, in which he not only         minefield was geographically
algorithms, in which a learning machine actively        covered some serious technical material with            oriented, so choosing the best
selects unlabelled datapoints from the data set and     admirable clarity, but he also managed to (almost)      direction to get her out of
requests their labels. He showed how this can           seamlessly integrate a collection of randomly           danger quickly was not going
improve speed of learning in many tasks, and            chosen words (Zephyr, Anaconda and Afghanistan)         to be straightforward (e.g.
outlined work on extending the principle to             into his talk, for which he deservedly received a       walking due North until you
regression, using Gaussian Processes.                   mini-Mexican wave from the front row!                   cross the edge of the strip is
    After a break for coffee, Mike Davies (Queen            All in all, an excellent meeting. Thought           not going to work if you’re
Mary, University of London), presented an overview      provoking and entertaining, and as ever an excellent    unlucky enough to find the
of ongoing work in the area of Audio Source             opportunity to discuss natural computing and            minefield runs North-South).
Separation and ICA. He outlined methods for blind       similar issues with friends and colleagues in a         On top of this, the torch could
source separation, where two audio sources              relaxed environment. Well done to everyone who          only illuminate up to six feet
                                                                                                                away so she was going to have
recorded by an array of sensors can be separated        contributed to the event.
                                                                                                                to get very close to the edge to
without any knowledge of the source characteristics.
                                                                                                                spot it.
The technique employed a frequency domain               See you in Cambridge!                                      “Hey, Henri, what’s the
framework and Mike explained how to get over the                                                                French for BA-BA-BOOM?”
problem of selecting the right ordering of sub-         Richard Ellis                                           said the pilot as the chopper
solutions (the ‘permutation’ solution). A practical     Stratum and BCS SGAI                                    dusted off, leaving Lisa to
                                                                                                                contemplate her options.
                           DIARY DATES 2003/2004                                                                Choosing to minimise the
                                                                                                                total number of yards she
2003                                                    8 December ASPGP 2003 1st Asia-Pacific Workshop         would have to walk in the
3 – 4 September NCAF – Human Computer                   on Genetic Programming Canberra, Australia. For         minefield in the worst-case
Interaction, Cambridge University (sponsored by         more information: http://evonet.dcs.      scenario, she set off on her
Microsoft Research, Cambridge).                         evoweb/news_events/conferences/conf945.html             optimal route.
For more information contact: Mark Cheeseman            14 – 17 December HIS 2003 The Third
                                                                                                                What is Lisa’s walking strategy and
e-mail:,                          International conference on Hybrid Intelligent
                                                                                                                how many yards at most would she
Tel: +44 (0) 1332 246989                                Systems, Melbourne, Australia. For more                 have to cover in order to guarantee
                                                        information:     leaving the minefield?
26 – 30 September FEA 2003 The Fifth
International Workshop on Frontiers in
                                                        2004                                                    The answers will be given at
Evolutionary Algorithms, Cary, North Carolina,
                                                        7 – 8 January (provisional date) NCAF –                 the next NCAF meeting
USA. For more information:
                                                        Fundamentals of Natural Computing.                      (3–4 September 2003,                                                              Fitzwilliam College,
                                                        For more information: contact Mark Cheeseman
conferences/conf935.html                                                                                        Cambridge).
13 – 17 October IAT 2003 The 2003 IEEE/WIC              Tel: +44 (0) 1332 246989
International Conference on Intelligent Agent           14 – 17 April ICEIS 6th International Conference        Fenella the Rottweiler
Technology, Beijing, China. For more information:       on Enterprise Information Systems, Porto,      Portugal. For more information:
6 – 8 November ICCI 2003 International Congress
                                                        20 – 22 April ACDM 6th Evolutionary / Adaptive
on Computational Intelligence, Medellín,
                                                        Computing in Design and Manufacture,University
Colombia. For more information:
                                                        of the West of England, UK. For more information:
                                                        conferences/conf946.html                                  Natural Computing Applications   Forum
COMMITTEE                       Researching complexity in clusters
  NOTES                               s part of the Novel Computing Initiative for

                                                                                                        Research Challenges
                                      2003/04, EPSRC recently awarded funding to 16
                                      research clusters in the general area of             One of the main difficulties in analysing large
Chairman                                                                                   interacting systems is how to extract useful top-
Graham Hesketh                  complexity. One of these research clusters, ‘Inference
                                and co-operative behaviour in large interacting            level information from the deluge of microscopic
Rolls-Royce plc                                                                            data, which tends to be highly fluctuating and very
                                systems’, is led by the Neural Computing Research
                                Group, Aston University. Members of the research           sensitive to changes in local conditions. Using
Secretary                       cluster include researchers from Birmingham                simulations and/or algorithmic solutions of such
Mark Cheeseman                  University, Btexact, King’s College London, Oxford         systems is infeasible as there are no practical
Rolls-Royce plc                 University, Reading University / JCMM (Met Office),        deterministic methods to search solution spaces
                                Royal Holloway, Tokyo Institute of Technology and the      that grow exponentially with the system size.
                                University of Florida.                                          We anticipate that our research programme will
                                                                                           focus on macroscopic and mesoscopic methods as
Dr Ian Nabney                                      Background                              they provide a top-level understanding of the system’s
Aston University                                                                           behaviour and can identify the emergence of global
                                Understanding the behaviour of large complex non-
                                linear systems is of great importance to the analysis      characteristics. Practical inference methods will also
Editor of Neural Computing                                                                 be devised based on the insight gained from the
                                and handling of information in modern society.
& Applications Journal                                                                     macroscopic analysis; these are essential for
                                These systems are characterised by a synergetic
Professor John MacIntyre        emergent behaviour of a different nature to that of a      obtaining specific solutions in complex systems, in
University of Sunderland        similarly large but uncoupled or linearly interacting      the absence of feasible algorithmic approaches.
                                system. Due to their complexity, especially the type            The main objective of the cluster is to stimulate
Managing Editor of              and number of interactions between components,             interdisciplinary research in the general area of
                                deterministic approaches to analysing such systems         inference and co-operative behaviour in large
                                are doomed; all feasible approaches rely on a              complex systems. This will be achieved through
Inspector Rick Adderley                                                                    interaction between theoreticians, researchers and
West Midlands Police            probabilistic definition of the problem using
                                techniques from statistics and statistical physics.        end users, leading to an integrated and coherent
                                     Very large complex systems are characterised by       research effort. The challenges facing the research
Publicity Officer                                                                          cluster, beyond integration, come at several domains:
                                multiple time and resolution scales. Microscopic
Richard Ellis                                                                              (i) setting up generic theoretical frameworks for the
                                interacting objects self-organise to construct
Stratum Management Ltd          mesoscopic weakly interacting agents, combined with        various problems; (ii) carrying out a full analysis of the
                                order parameters which lead to a rich diversity of         generic models; (iii) exploiting the insight gained from
Mandy Bradley                   macroscopically observable phenomena, such as              the analysis for devising feasible inference methods;
Technical Forecasts Ltd         coherence, phase transitions (drastic changes in           (iv) employing the inference methods for realistic
                                macroscopic behaviour at specific critical values of       large scale applications.
                                the model’s parameters), and structural chaos. These            Our strength is in the generic nature of our
Dr Richard Everson                                                                         analytic approach, which is not limited to one
University of Exeter            systems are driven by non-linearity and, very often,
                                stochasticity; the latter comes from the natural           domain, but can be employed to study realistic large
                                characteristics of the problem, inherent uncertainties     scale interacting systems of many types. Therefore
Dr Iead Rezek                                                                              we propose to develop our research programme
                                and statistical behaviour, as well as from the practical
Oxford University                                                                          across three broad domains, which will then
                                difficulties of modelling and extracting information at
                                the microscopic level. By extending techniques from        motivate the specific projects. These broad domains
’Bert Bullen                    statistical physics, synergetics, many-body theory,        are: microscopic inference in large connected
BAE Systems                     information theory and pattern analysis we wish to         systems (weather, environment and data mining),
                                address issues in the behaviour of complex systems at      distributed communications and computing
                                each of these resolution scales.                           networks and multi-agent system dynamics (from
Dr Elaine Martin
                                     This research cluster brings together expertise       ultra devices to the immune system).
University of Newcastle
                                from a range of disciplines in order to develop a road                    Planned Activities
Please contact NCAF             map for research in this area and generate a research
                                programme to address the following issues:                 The planned activities aim to join theoreticians,
through Mark Cheeseman                                                                     application developers and end users to forge a
                                ● devising feasible inference methods for very
Secretary – NCAF                                                                           consensus on the best directions for further
                                     large connected systems;
PO Box 5944                     ● analysing typical system behaviour in specific
                                                                                           research in order to overcome the challenges in this
Derby DE24 8ZD U.K.                  instances of large complex systems under given        domain. In the six-month course of the programme
Tel: +44 (0) 1332 246989             disorder (e.g., topological constraints) and a        we shall run activities to gather information and
Fax: +44 (0) 1332 247129             dynamically changing environment;                     ideas, analyse and evaluate them in open forum,
                                ● using the insight gained from the analysis to
                                                                                           and then disseminate the results back to the
                                     support local decision making as a method for         research and user communities.                                                                         The main activities of the cluster will be two
                                     controlling and directing global network
                                     properties                                            workshops: a kick-off meeting held on the 17 July
                                     Our long term aim is to move from simulation          and a second meeting, open for all, towards the
                                and modelling, which are typically infeasible in large     middle of October.
                                complex systems, to the extraction of higher level             We see outreach and community-building as
     Edited and Produced by:                                                               very important components of this project, and
        Chris Hawthorne         variables that allow us to monitor, predict and
    Forum Communications        design such systems. We suggest the exploitation of        intend using both active and passive activities to
        Eastgate House          similarities between many-agent systems such as            promote wider involvement. We would like to get in
         Eastgate Street        nervous and immune systems, environmental                  touch with end users to find out their needs and
     Winchester SO23 8DZ                                                                   interests to help direct our research towards useful
       Tel: 01962 877833        models, future Internet, ultra-device chips and
                                communication networks and develop principled              long-term goals. We would like to invite interested
       Fax: 01962 877988                                                                   NCAF members to participate in our activities and
             e-mail:            information processing approaches to describe the   resultant behaviour, so that we can design and             to let us know of their interests.
                                forecast better for the future.                            David Saad Aston University
Review of Cambridge                                                MEMBERS’ NEWS AND VIEWS
Meeting                         Deadline for contributions for the next edition – 6 October 2003. Please send to: Managing Editor –
Preview of Winter Meeting       Inspector Rick Adderley, e-mail: or