D SLEEMAN, Z LUO, G CHRISTIE, G COGHILL (2004) Analysing Time Series Medical Data-sets

Proceedings of Knowledge Based Systems & Services for Health Care. Bonn: May 2004, p 1-4.

Abstract At the 1999 AIME Conference [1] we reported a decision tree study on a subset of data, including time series physiological data, admissions characteristics, and outcome after traumatic head injury. That study analysed total duration for which patients had, for example, raised Intracranial Pressure, but it did not consider temporal relationships between various physiological and clinical events. In this study, we have addressed that issue in a number of ways. Firstly, by using a workbench, AAB, to display the Real Time data-set and asking clinicians to make predictions of expected outcome based on complete physiological and clinical data. Secondly, repeating the exercise with a reduced (more compact) representation for the physiological data. Thirdly, patterns were generated, including “adjacent” physiological parameters and clinicians were asked if they are likely/very unlikely to cause a particular major physiological event or outcome. Finally, we implemented a module to test patterns of the form:

IF X happens then Y will happen between T1-T2

against patient time-series data. Results of all these studies have so far not been conclusive [2]; it has been suggested that the brain is currently not very well understood physiologically, and that a similar set of analyses should be applied to a simpler organ. Given that significant amounts of data are now available for patients undergoing dialysis, we have chosen to do an analogous study in this area; also the physiology of the renal system is much better understood. We have outlined some additional studies we plan to undertake using data-mining, theory refinement and knowledge base refinement approaches. (Download)

D W FOWLER, D H SLEEMAN, G WILLS, T LYON & D KNOTT (2004) The Designers' Workbench: Using Ontologies and Constraints for Configuration

Technical Report AUCS/TR0406, Department of Computing Science, The University of Aberdeen.

Abstract Typically, complex engineering artifacts are designed by teams who may not all be located in the same building or even city. Additionally, besides having to design a

part of an artifact to be consistent with the specification, it must also be consistent with the company’s design standards.

The Designers’ Workbench supports designers by checking that their configurations satisfy both physical and organisational constraints. The system uses an

ontology to describe the available elements in a configuration task. Configurations are composed of features, which can be geometric or nongeometric, physical or

abstract. Designers can select a class of feature (e.g. Bolt) from the ontology, and add an instance of that class (e.g. a particular bolt) to their configuration. Properties

of the instance can express the parameters of the feature (e.g. the size of the bolt), and also describe connections to other features (e.g. what parts the bolt is used to hold together). (Download)

T NORDLANDER, K BROWN, & D SLEEMAN (2003) Identifying inconsistent CSPs by Relaxation. Technical Report AUCS/TR0304, Department of Computing Science, The University of Aberdeen.

Abstract: How do we identify inconsistent CSPs quickly? This paper presents relaxation as one possible method; showing how we can generate relaxed CSPs which are easier to prove inconsistent. We examine different relaxation strategies based on constraint graph properties, and we show that removing constraints of low tightness is an efficient strategy which is also simple to implement.

(Download)

D. SLEEMAN  & S WHITE (2002). Technical Report AUCS/TR0202, Department of Computing Science, The University of Aberdeen.

Abstract  This paper introduces a stand-alone case/knowledge acquisition tool, called COCKATOO (Constraint-Capable Knowledge Acquisition Tool), which uses an Extended BNF grammar to represent the main characteristics of the (domain) cases. Further, we also took the opportunity to build a tool that is both more flexible and powerful by augmenting the context-free grammars with the expressiveness of constraints. COCKATOO was implemented using the SCREAMER+ declarative constraints package. Additionally, the paper discusses several uses of the tool by both the developers and, more significantly, by a group of knowledge engineers.

A. HAMEED & D. SLEEMAN (2000). Technical Report AUCS/TR0001, Department of Computing Science, The University of Aberdeen.

Abstract  We present the initial results from a knowledge elicitation exercise carried out with human experts in the domain of PC specification. The experts advise novice and proficient computer users and specify configurations for personal computers tailored to suit user needs. Using standard knowledge elicitation techniques such as semi-structured interviews and stimuli sets, we have been able to elicit domain terminology and concepts from each of the experts. We have also obtained an indication of the problem-solving strategies employed. The report contains a complete record of the knowledge elicitation sessions in the form of subjects' protocols which have been transcribed verbatim from audio tapes. These are followed by brief analyses. A comprehensive set of glossaries extracted from the expert protocols is given in the appendices. The resultant protocols and data sets are being analysed to construct ontologies of each expert's conceptual structures, the results of which will be published in a subsequent report.