John Washbrook
Emeritus Professor of Computer Science, University College London
Information
Email: J.Washbrook@cs.ucl.ac.uk
Click here for more information on UCL-CS.
Biography
I joined the University of London Institute of Computer Science as a lecturer in 1972. Upon the disbandment of the Institute I transferred to UCL, where I became a professor of Computer Science. I took the opportunity to retire early in 2002 but maintain my research collaboration with Professors Christine Hall, Elpida Keravnou, and Nigel Titchener-Hooker.
My first job after graduating from Durham with a Mathematics degree was at the de Havilland Aircraft Company (later Hawker-Siddeley Dynamics). Here, as a missile systems engineer, I was involved in electronic control systems using analog devices. I then joined Honeywell EDP on the sales support staff, involved in the design and implementation of (mainly) financial systems working with assembler and Cobol. Honeywell had a tape ‘polyphase sort’ which was a market leader at the time, and my interest in learning more about this and the Cobol compiler led me to take the conversion MSc in Computer Science at the University of London Institute of Computer Science (ULICS). This course still lives on as the MSc in Computer Science at UCL, and has provided many of my former and current colleagues.
Having gained the MSc, like many other graduates of the course, I had a change of direction. I joined the Medical Research Council’s Computer Services Centre as Deputy Director. My major project there was the creation of a knowledge-based system for the estimation of gene frequencies. Whilst I was there a position on the academic staff of ULICS became available and that led to another change, into lecturing and my current position. With Elpida Keravnou I became involved in Knowledge-Based Systems with biotechnologists (process design/business process re-engineering) and later with radiologists (temporal diagnosis of skeletal dysplasias and malformation syndromes).
Research
(See also Research Grants)
My work has had a consistently medical/biological bias. Before becoming a lecturer I worked for the Medical Research Council, in particular on the creation of a simple-minded knowledge-based system with Bayesian influences for the estimation of gene frequencies from blood group data, working with/for Dr A E Mourant. This was my introduction to representing expert knowledge.
Malcolm Lilly (Biochemical Engineering) made contact in 1986 and this led with Elpida Keravnou to some publications on representing knowledge about bioprocessor design. Nina Thornhill (E&E Engineering) introduced us to the qualitative concept of a process signature and that was one of the ideas we worked on in the EQUATOR (Environment for Qualitative and Temporal Reasoning) project.
About this time Elpida Keravnou and I were approached by two radiologists, Dr Richard Dawood and Professor Christine Hall from the Hospital for Sick Children, on the topic of the diagnosis of skeletal dysplasias and malformation syndromes. This is a difficult area as there are over 2000 syndromes, and their features can be quite subtle. The technical interests here were the separation of background knowledge from expert knowledge and the incorporation of temporal knowledge – when features appear can be as important as if they appear. We had two successful projects funded by the Leverhulme Trust. This work led to REAMS, a radiological electronic atlas of malformation syndromes. REAMS posed interesting challenges in the representation of mixed taxonomic and meronomic structures, which are still not resolved. Additionally it took me into HCI and databases. REAMS gained an award at RSNA ’99, which we are assured is no mean feat.
In all the work I have done in the medical domain I am of course interested in generic solutions with potential for much wider application, and it is important to me that they should be applied and essential that they should scale up.
More recently my connection with the biochemical engineers has revived following an approach from Nigel Titchener-Hooker of the Department of Biochemical Engineering, UCL. His pioneering work involves the integrated modelling of both business and technical decisions in the production of biopharmaceuticals (other approaches model only one aspect). An unusual aspect of this domain is the high level of risk and cost, which is matched by the huge potential profits. The success of the work has lead to the funding of a second large Innovative Manufacturing Initiative (IMI) research programme.
Teaching
During my working life I have used a number of languages – various assemblers, Cobol, PL/1, algol60, Fortran IV, Ratfor, C, Pascal, Visual Basic, and Java – as well as having had meaningful relationships with algol68, Lisp (Allegro and Scheme), Snobol/Spitbol, Simula and C++ (enough to help students). Actually I delayed getting to really know C++ long enough that (for teaching purposes here) it got overtaken by Java – a really successful justification of procrastination.
I have taught expert systems to final-year students and a variety of first- and second-year courses: programming courses in Ratfor, PL/C, Pascal, C, C++, and Java, computer systems architecture, systems software, simulation, data structures and algorithms, compilers, databases. I have also taught many of these topics to students on the conversion MSc course.
Over the years I have developed an approach based upon annotated syntax diagrams that in a few lectures introduces students to the principles of parsing and code generation in a practical way with strong theoretical underpinning. It is limited to control structures, with no coverage of functions/block structure. I see its value as introducing students to executable specifications/formal definitions in a very practical way. Many students these days come to us with no background in formal definitions and for them it is quite a culture shock. It is a valuable complement to their programming activities as it makes them think rather than hack. It also helps them understand high-level languages much better, interpret compiler error messages meaningfully, and demystifies the gap between high-level and low-level languages. Following the approach through syntax diagrams it is a relatively small step in the second year to introduce grammars, compiler-compilers, and shift-reduce parsing.
Links to: undergraduate programmes and taught masters programmes.
Administration
For all but a few of my 30 years of full-time employment at UCL I was heavily involved in the internal running of the department, as Admissions Tutor, Departmental Tutor, Director of the MSc conversion basic course, and Chair of Teaching Committee – except for when I was involved with EQUATOR. If this makes me look employable in some way I don’t think I really want to know about it.
Personal
My wife Fran and I have worn out three or more kettles. (Topical research shows that most kettles outlast marriages.) Fran has a fine arts background and teaches life drawing. We have four daughters who have 6 and a bit degrees between them and similar preferences to their parents for interesting rather than financially rewarding jobs – in medical authoring and editing, ecology, and museum and gallery management.
I enjoy playing Go. At one time I thought I was on the way to shodan, but my family life had more than enough competition from teaching, research, and (yes) administration. Go has taught me much about myself.
Also having known many Chinese students over the years, I am belatedly learning Mandarin. Being in the position of a student in a totally different and challenging environment is very enlightening.
Research Grants
Publications
Five selected publications with annotations
Hall C.M.,
Washbrook J. (2000) “Radiological Electronic Atlas of Malformation Syndromes
and Skeletal Dysplasias (REAMS)”, ISBN 0 19 268593 7, Oxford University Press.
REAMS is a radiological atlas consisting of over 7 000 X-ray images with some 20 000 findings covering over 200 conditions. It is presented on a CD with a sophisticated search engine that allows images to be selected in a variety of ways, at differing levels of detail of finding. This is the first electronic version of an atlas at this scale; the nearest quality ‘competitor’ is a conventional (book) atlas dating from 1985 which, being paper-based, has far fewer images and coverage – and of course is not interactive. REAMS is almost certain to make at least as lasting a contribution as a diagnostic and reference tool as its book predecessor if only because of the enormous effort involved in its creation and the scarcity of such extensive and authoritative expertise as Hall’s. For her, it is her legacy to the subject. For me as the computer scientist it presented a number of fascinating challenges of which the greatest were representing and retrieving knowledge at arbitrary levels of detail in a mixed meronomic and taxonomic structure, and in translating encoded findings into the language that radiologists use. There is more to be written about the mixed structures when I have really understood the solution better. The translation of findings I found surprisingly difficult, and I almost agreed at one stage with others who had thought it impossible – but if a human can do it, so can a computer. REAMS was awarded a certificate of merit at RSNA ’99, the Radiological Society of North America’s convention.
Farid, S., Novais, J.L., Karri,
S., Washbrook, J., Titchener-Hooker, N.J. (2000) A tool for modelling strategic
decisions in cell culture manufacturing Biotechnology Progress, 16, 829-836.
This paper with three research students presents the development of a software tool for harnessing business and process considerations in decision-making within biopharmaceutical manufacture. It is the result of collaborative research between the Department of Biochemical Engineering and the Department of Computer Science, and arises from an IMI grant in the area of business process modelling. The paper demonstrates the particular insight into the inter-play between process and business factors that shapes any development.
Keravnou,
E. and Washbrook, J. (2001) “Abductive Diagnosis using Time Objects: Criteria
for the Evaluation of Solutions”, Computational Intelligence, 17(1),
87-131.
The novel work described here
draws upon the experiences of working with diagnosticians at Gt. Ormond
St. Expert diagnosticians employ
different, complex criteria at different stages of their reasoning. These complex criteria are composed from
more primitive criteria, which therefore need to be modelled. In abductive diagnostic reasoning the
proposal of solutions and their evaluation are closely coupled. The paper presents an abductive framework
for diagnosis.
Time can be of great significance: when a finding presents, and for how long,
can be just as important as if the finding presents. The modelling of time is difficult and has largely been ignored
by others. In this work we introduce
time-objects, recursively-structured objects that in this context correspond to
complex findings with lifetimes. The
integration of time with primitive criteria in diagnosis is illustrated by
reference to our Skeletal Dysplasias Diagnostician system.
Keravnou E.T., Dams F.,
Washbrook J., Dawood R.M, Hall C.M., Shaw D. (1992) “Background Knowledge in
Diagnosis", Artificial Intelligence in Medicine, 4(4),
pp. 263-279.
One of our goals as computer scientists in our diagnostic reasoning work was to separate out general background knowledge from expert knowledge. For example, there is knowledge that any registrar could be expected to have about anatomy and development, and the knowledge (factual as well as diagnostic) that makes an expert an expert in a particular speciality. The separation of knowledge would allow the background knowledge to be available to other areas of expertise in related domains, and make the expert knowledge more ‘compact’ and thus more accessible for study. It turns out that this separation is much more difficult than we expected, indeed it may be impossible, as experts actually have a different version of the background knowledge. Thus the task of representing expert knowledge also involves representing these differences. There is more to be written about this
Washbrook
J. and Keravnou E.T. (1990) “Making Deepness Explicit", Artificial
Intelligence in Medicine, 2(3), pp.129-134.
This is one of my favourite papers as it explores what is meant by the use of the word ‘deep’. Appreciating the derivation and meaning of words is vital to clear thinking and communication. When I entered into this field it took me quite a while to unravel what was variously meant by a ‘deep’ model. This paper presents a working definition of ‘deepness’.
Chapter in Book
Washbrook J. (1976) “The Estimation of Gene Frequencies by Computer", Chapter 13 in Mourant A.E. et al ‘The Distribution of Blood Groups and Other Polymorphisms', Oxford University Press.