DEPARTMENTAL SEMINAR (COMPUTER & COGNITIVE SCIENCES) SPRING TERM 1999 TIME AND PLACE: Thursdays 4pm (Unless otherwise stated) Lecture Room 7, School of Computer Science ORGANISER: Marta Kwiatkowska (M.Z.Kwiatkowska@cs.bham.ac.uk, ext 7264) Sammie Snow (S.Snow@cs.bham.ac.uk, ext 4774) SPEAKERS and TITLES -- Thu 21 Jan Prof. Allan Ramsay, UMIST Constructing Discourse Models -- Fri 22 Jan ***4pm, LR7*** Dr Allen Long, City University Bankruptcy Prediction Models: Probabilistic Neural Networks versus Discriminant Analysis and Backpropagation Neural Networks -- Thu 28 Jan ***Note change of speaker*** Dr Eike Ritter, School of Computer Science, University of Birmingham Abstract Machines for Functional Languages -- Thu 4 Feb Prof. Rae Mackay, Earth Sciences, Univ of Birmingham A Brief Portrait of Hydrogeological Simulation -- Thu 11 Feb Prof. Tom Maibaum, Imperial College Systematising reactive system design -- Thu 18 Feb Prof. Gregory McCulloch, Department of Philosophy, University of Birmingham The Demonic Dilemma -- Thu 25 Feb Dr Luca Cardelli, Microsoft Research, Cambridge Abstractions for Mobile Computation -- Thu 4 Mar Prof. Martin Russell, Electronic & Electr. Engineering, University of Birmingham The good, the bad and the ugly : a look at the state-of-the-art in automatic speech recognition -- Tues 9 Mar * SPECIAL SCIENCE FACULTY LECTURE * TUESDAY, NOT THURSDAY Prof Robin Milner, FRS, University of Cambridge Tuesday 9th March 5.15. Avon Room, University Centre Computing is Interaction -- Thu 11 Mar Dr Muffy Calder, University of Glasgow Analysing a Basic Call Protocol using Promela and XSpin -- Fri 12 Mar ***4pm, Note different day*** Sarbjit Sarkaria Notes on Constructing a Large Distributed System -- Thu 18 Mar Geoffrey Hall, Nortel Networks, Vice President International Carrier Solutions Technology Telecommunication Network Evolution -- Thu 25 Mar Dr Jim Woodcock, University of Oxford Industrial-Strength Refinement ABSTRACTS ----------------------------------------------------------------------------- -- Thu 21 Jan Prof. Allan Ramsay, UMIST Constructing Discourse Models Most current theories of natural language semantics take the view that natural language is inherently `dynamic' -- that when a speaker produces an utterance and a hearer receives it they are both changed by it. The current talk will present a concrete model of this process: the key notion is that the hearer's task is to construct a model which admits the content of the utterance. This is done by adding the interpretation of the utterance to the current discourse and exploring the resulting tableau. Any open branch in the tableau characterises a partial model: the first such model that is found is taken to be the new discourse state. This approach has a number of demonstrable benefits in dealing with various key problems in natural language semantics, notably in dealing with reference and with underspecification. The talk will also touch on some more speculative work on extracting discourse relations by this means. ----------------------------------------------------------------------------- -- Fri 22 Jan ***4pm, LR7*** Dr Allen Long, City University Bankruptcy Prediction Models: Probabilistic Neural Networks versus Discriminant Analysis and Backpropagation Neural Networks In this talk we present probabilistic neural networks (PNN) as an alternative quantitative technique to both linear discriminant analysis (LDA) and backpropagated neural networks (BPNN) for forecasting corporate solvency. Although traditionally this task has been approached with rather simpler linear techniques such as LDA, there is increasing empirical evidence of the superiority of BPNN models to LDA in terms of its ability to accurately forecast corporate financial health. However, some recent work seems to indicate the superiority in forecasting performance displayed by BPNN may not be as significant as first thought. This paper suggests an alternative technique, probabilistic neural networks, for the assessment of corporate financial distress which can produce superior performance to both LDA and BPNN's while not suffering from the shortcomings found in BPNN's. ----------------------------------------------------------------------------- -- Thu 28 Jan ***Note change of speaker*** Dr Eike Ritter, School of Computer Science, University of Birmingham Abstract Machines for Functional Languages Implementations of functional languages are very often described in terms of abstract machines. In this talk I will start by describing the ideas behind commonly used abstract machines and explain the two paradigms, namely graph reduction and environment machines. In the second half I will focus on our work, where we use linear logic to improve memory management of abstract machines. ----------------------------------------------------------------------------- -- Thu 4 Feb Prof. Rae Mackay, Earth Sciences, Univ of Birmingham A Brief Portrait of Hydrogeological Simulation Understanding the movement of water and fate of chemicals in geological systems is a key issue for environmental protection and the development of water resources and waste disposal. Computer based simulation methods have played, and will continue to play, an important role in developing this understanding. Advances in the power of computers have lead to the rapid evolution of increasingly sophisticated tools for predicting the impacts of future changes on the subsurface environment and on risks to human health. However, more complex simulation tools are not always associated with better solutions to problems involving natural systems. Indeed, models can be wildly inaccurate. The hydrogeologist's art is to infill the many missing pieces of the jigsaw of geometric, process and property interactions that characterise a hydrogeological system. Harnessing the power of computing technology to assist this process provides valuable help. This presentation will provide a brief overview of hydogeological simulation and some recent developments. ----------------------------------------------------------------------------- -- Thu 11 Feb Prof. Tom Maibaum, Imperial College Systematising reactive system design Reactive systems are a very important class of systems for all engineering organisations. The role of software is now dominant and its shre of system construction costs is ever increasing. However, in spite of much research devoted to reactive system development, proper engineering languages, methofds and tools as construed by the conventional engineering community, are not available to support the technology. We examine the state of play and suggest a way ahead for putting into place appropriate technology for industry. Software architecture, design by classification and the role of conventional design languages play a role, along with appropriate scientific/mathematical underpinnings. ----------------------------------------------------------------------------- -- Thu 18 Feb Prof. Gregory McCulloch, Department of Philosophy, University of Birmingham The Demonic Dilemma The real problem for Descartes in the First Meditation is not knowledge but intentionality itself: not how we could know about the world but how we could even think about it. This claim is briefly defended then the moral is applied to various kinds of brains in vats. I hope to make palatable the conclusion that they have no mental life worthy of the name. ----------------------------------------------------------------------------- -- Thu 25 Feb Dr Luca Cardelli, Microsoft Research, Cambridge Abstractions for Mobile Computation (Joint work with Andrew D. Gordon) There are two distinct areas of work in mobility: "mobile computing", concerning computation that is carried out in mobile devices, and "mobile computation", concerning mobile code that moves between devices. These distinctions are destined to vanish. We aim to describe all aspects of mobility within a single framework that encompasses mobile agents, the ambients where agents interact and the mobility of the ambients themselves. The main difficulty with mobile computation is not in mobility per se, but in the crossing of administrative domains. Mobile programs must be equipped to navigate a hierarchy of domains, at every step obtaining authorization to move further. Therefore, at the most fundamental level we need to capture notions of locations, of mobility and of authorization to move. We identify "mobile ambients" as a fundamental abstraction that generalizes both dynamic agents and the static domains they must cross. From a formal point of view we develop a simple but computationally powerful calculus that directly embodies domains and mobility (and little else). The calculus forms the basis of a small-language/ Java-library. We demonstrate the expressiveness of the approach by a series of examples, including showing how a notion such as "crossing a firewall" has a direct and analyzable interpretation. ----------------------------------------------------------------------------- -- Thu 4 Mar Prof. Martin Russell, Electronic & Electr. Engineering, University of Birmingham The good, the bad and the ugly : a look at the state-of-the-art in automatic speech recognition In 1949, George Orwell predicted that by 1984 the pen would be "an archaic instrument, seldom used even for signatures" and that, apart from short notes, it would be "usual to dictate everything in to the speak-write". Now, in 1999 it looks as if Orwell's prediction may finally be coming true. All of the popular computing magazines, and even the television, carry advertisements for automatic dictation systems. Portable phones incorporating speech recognition are also on the market, and the technology is likely to appear in a number of other applications in the near future. So, is automatic speech recognition a solved problem? Certainly, huge progress has been made over the past 20 years. However, in the full context of spoken language processing by computer, I shall argue that the answer to this question is an emphatic 'no'! Although significant progress has been made in terms of recognition accuracy, this has resulted from incremental development of a statistical approach, hidden Markov modelling, which has been applied to speech recognition for at least the past 25 years. These advances in performance have not been matched by fundamental progress in our understanding of how to model speech patterns or how to incorporate descriptive speech 'knowledge' as computationally useful constraints in the speech recognition process. The talk will begin with a brief discussion of the factors which make speech recognition such a difficult pattern processing problem. This will be followed by an overview of the state-of-the-art in automatic speech recognition, in terms of performance and speech pattern modelling. Then I will argue that although current approaches have severe limitations in the full context of speech pattern modelling, and many of the assumptions are clearly flawed, there also very positive aspects which one would want to preserve in future systems. ----------------------------------------------------------------------------- -- Tues 9 Mar * SPECIAL SCIENCE FACULTY LECTURE * TUESDAY, NOT THURSDAY Prof Robin Milner, FRS, University of Cambridge Tuesday 9th March 5.15. Avon Room, University Centre, The University of Birmingham. Computing is Interaction Traditionally, computing has been overwhelmingly concerned with the notion of *algorithm*; something which finds the answer to a well-defined question and delivers it. Moreover, early computer science was dominated by the idiom of the *sequential* algorithm. With the advent of multiprocessors, programming languages began to require more; they were extended to allow concurrent activity. But today's concurrent languages tend to preserve the sequential idiom, providing a top-dressing of concurrency and interaction. But it is ugly and redundant to build a science of computation in this stratified manner. Besides, computer science -or informatics- is no longer only concerned with *prescribing* what will happen in the small world of a sequential computer; it must also *describe* what happens at large in parallel computers, in networks, in hybrid human-machine systems, and even in nature. Like energy, information flows everywhere all at once. The thesis of this lecture is that we can base all computation -both prescriptive and descriptive- upon an atomic notion of interaction. In this model, a computational entity -- a process, object or agent -- is nothing more than its ability to exchange information with others, just as physical entities exchange energy. The analogy goes further: informatic agents (eg on the internet) are mobile, just as physical agents are. We need a theory to understand them. I shall give simple illustrations, using a calculus of mobile interactive processes known as the pi calculus, which makes a first step towards such a theory. ----------------------------------------------------------------------------- -- Thu 11 Mar Dr Muffy Calder, University of Glasgow Analysing a Basic Call Protocol using Promela and XSpin It is difficult to find formalisms for modelling telecommunications services which offer both the appropriate language constructs and powerful automated reasoning tools to aid analysis. In this talk I will consider how Promela (process meta language) and the model checker XSpin might meet some of the requirements. A network of basic call models, with a simple asynchronous communication protocol, will be developed and then analysed using a variety of techniques, including model checking for linear temporal logic formulae using XSpin. The results are a little surprising! No prior knowledge of Promela and XSpin will be assumed, and during the talk I will indicate how this work fits in with other research at Glasgow in the area of feature interaction analysis and telecommunications software. ----------------------------------------------------------------------------- -- Fri 12 Mar ***4pm, Note different day*** Sarbjit Sarkaria Notes on Constructing a Large Distributed System This talk will interest anyone who would like to know more about practical software engineering and what\271s involved in building large distributed systems. The talk will discuss the role of object oriented design versus waterfall models, development of a software architecture necessary for distributed systems and how a well defined development process can lead to improved software quality. The contents are based on over six years of experience gained during the construction of a large air traffic control (ATC) system for the Canadian government and will include examples of problems specific to ATC. ----------------------------------------------------------------------------- -- Thu 18 Mar Geoffrey Hall, Nortel Networks, Vice President International Carrier Solutions Technology Telecommunication Network Evolution The rapid growth in data traffic driven by Internet usage and corporate communications is changing the principles by which Network Operators define networks. This has generated industry debate on future architectures and the standards and technologies needed to make them successful. The seminar will provide an introduction to the changes in Network architecture and the challenges it poses for product and software developers. No prior knowledge of Telecoms is assumed. --- About the speaker: Responsibilities: Managing Director of the Maidenhead Research and Development Lab which produces software for a range of Telecommunications products for Network Operators in Europe and the Asia-Pacific region. Background: A graduate of Electronic Engineering has spent 23 years in the Telecommunications industry with ITT, STC and Nortel Networks as a development engineer and more recently in the management of product development. ----------------------------------------------------------------------------- -- Thu 25 Mar Dr Jim Woodcock, University of Oxford Industrial-Strength Refinement Research in software engineering is characterised by the interplay between theory and practice: the best research finds its direct application in industry, whilst the best industrial practice provides interesting problems for the researcher. In this talk, we describe an industrial application as a case study in the introduction of new techniques in industry. It concerns using advanced software engineering methods to make money: literally. The project involves developing a smart-card for electronic commerce; this is a major new product and a state-of-the-art application. Formal methods were used very successfully to model behaviour to guarantee security, with the product being the first to gain certification to ITSEC Level 6. The mathematical work was completed under budget and well within the time allowed. This was a real product which had to hit the streets, but the client wanted proof of security. We discuss the interesting theoretical problems that we encountered, the tension between industrial and academic concerns, and the way that such projects stimulate software engineering research. -----------------------------------------------------------------------------