We describe recent work on image processing aimed at early diagnosis of breast cancer. Algorithms for removing degrading imaging effects, for eliminating scatter, and for identifying various sorts of breast tissue are reported. We discuss the development of a neural network to find masses, that correspond to potential cancers. Work aimed at combining x-ray mammography with Magnetic Resonance Imagery for early detection of breast cancer in younger women will be reported.
As speech archives grow, effective methods of retrieving individual spoken `documents' are required. The talk will describe collaborative work involving the Computer Laboratory, Engineering Department and Olivetti Research Limited initially aimed at retrieving video mail messages, but developing general techniques suited to other applications eg news retrieval. The research combines statistically-based retrieval techniques hitherto used for text with established speech recognition methods, seeking for adapt these to speech retrieval conditions including recording quality, informal material, unpredicted user needs, convenient output presentation, integration of retrieval with other workstation operations and, eventually, joint use of audio and visual cues in multimedia retrieval. The seminar will focus on the methods used, test material, and experimental results to date.
Developments in information technology are so rapid that most citizens and most legislators are unaware of the capabilities of the technology. Yet many proposed systems (road pricing, for example) could damage our quality of life if they are implemented naively or incompetently. This seminar discusses the proposition that the arguments for planning controls in physical space apply with equal force to cyberspace, and that we, the experts, have a duty to our fellow citizens that we are failing to discharge.
Object-oriented programming languages have received increasing attention over the past decade. Two important features of these languages are subtyping, which allows objects of one type to be used in place of objects of another, and inheritance, which allows significant code reuse during program development. This talk will summarize aspects of object-oriented languages that appear to be important in practice and discuss several programming language design problems. These included the choice to separate implementations from interfaces, which is useful in system design but raises problems with the implementation of typed binary operations, and the relationship between subtyping and inheritance. Time permitting, some type-theoretic models of object-oriented languages will be presented and compared.
In this talk I will survey the work of the Leeds' Qualtitative Spatial Reasoning Group. I will describe the First Order Formalism which provides a calculus for representing and reasoning about spatial regions in an analogous manner to temporal interval logic. Convex and various kinds of concave spatial regions are handled. Mechanisms for reasoning about spatial regions over time will be described. Applications of the calculus in Geographical Information Systems, Visual Programming Languages and Reasoning about Physical Systems will be surveyed. I will also show how much of this apparatus can be reexpressed in Intutitionistic Propositional Logic to computational advantage.
A property is a relationship between the input events supplied to a component and the output events it produces. Properties pertain to the set of all possible valid event sequences that apply to the component in question, and usually are determined with the component in isolation. When two such components are legitimately interconnected, it is normally the case that the resulting system will not preserve the individual component properties, even in the case where all interconnected components exhibit the same property. New properties, not present in some of the interconnected components, may emerge on interconnection.
Applications in security, fault tolerance and faithful execution, amongst others, would benefit from a certainty that when components that enforce a desirable property are interconnected, the result also exhibits that property. The problem is one of very long standing.
Recent encourageing research results will be reviewed in this talk. It will be shown that there exist practical component conditions that always allow property-preserving composition, and that the composed systems can be treated hierarchically in further composition. Finally, some outstanding difficulties will be presented.
The Standard ML language incorporates a module system that provides flexibility through a novel form of parameterized module called a functor. The original design admitted only "first-order" functors whose parameters and results were simple modules. More recently, this design has been extended to a truly higher-order module system by allowing functors to have functors in their parameters and results. This extension has brought into focus the central issue of how type information is propagated through functors. I will illustrate one approach to type propagation, through an operational semantics for higher-order functors and an new implementation closely modeled on this semantics. I will conclude by discussing alternative designs and possible future evolution of the ML module system.
Open hypermedia is a term used by the hypermedia research community to describe hypermedia systems in which the hypermedia links are themselves regrarded as objects, stored separately from documents (which can be maintained in their native formats). The World Wide Web in its common use is a closed hypermedia system because link information is embedded within HTML documents. This talk presents the open hypermedia philosophy, introduces the link service abstraction and discusses its implementation within a distributed hypermedia architecture. Early experiences of a distributed link service will be presented, together with an overview of an interpreted language for describing units of publication which associates the notion of a continuation with a hypertext link.