Machine Learning, Knowledge Acquisition
& Refinement Research at Aberdeen

Work at Aberdeen encompasses a broad range of activities: cooperative knowledge acquisition and refinement; machine discovery; nearest-neighbour learning & feature selection; and learning in the context of software agents.

This page summarises recent work and lists external publications and theses. For further information contact:

• Professor Derek Sleeman
  sleeman@csd.abdn.ac.uk

• Dr Pete Edwards
  pedwards@csd.abdn.ac.uk

Activities - Past, Present, Future

A program to be launched in the UK in 2000.

The UK Engineering & Physical Sciences Research Council (EPSRC) has recently established several IRCs (Interdisciplinary Research Collaborations) in IT. The AKT (Advanced Knowledge Technologies) IRC is to be funded for 6 years from the 1st of October 2000, is to be co-ordinated by Nigel Shadbolt (Southampton), and involves Aberdeen (Derek Sleeman), Edinburgh (Dave Robertson & Austin Tate), the Open University (Enrico Motta & Marc Eisenstadt) and Sheffield (Yorick Wilks).

In an information society, it is essential to possess, or to be able to access, knowledge in order to make sense of information, function efficiently, and provide good products and good service. Yet there are fundamental problems associated with the management of knowledge (capturing, representing, and distributing knowledge) that for too long have been addressed in isolation. The IRC will develop and extend a range of technologies to provide integrated methods and services for the acquisition, modelling, reuse, retrieval, publishing and maintenance of knowledge. It will bring these technologies together in a unified context that incorporates state of the art communication technology to enable rapid, wide, efficient distribution of key knowledge assets. The IRC will undertake fundamental research in particular knowledge technologies but it will also bring together relevant work and produce practical results. The proposal brings together a strong set of highly respected universities with complementary expertise and experience, across a wide range of disciplines, in dealing with knowledge-intensive methodologies and technologies. It has attracted a significant and enthusiastic response from industry, and the close involvement of several key commercial and industrial organizations that will ensure the usability of the results. We believe this IRC will provide timely, exciting and necessary support to the growth of the knowledge economy.

Total funding for the IRC will be ~£7 million; Aberdeen's share will be ~£1.2 million and will support three research fellows, support staff, and some research students.

Recent work in this area has been carried out across three sub-themes:

• Enhancing the efficiency of an earlier rule refinement system - leading to the STALKER system.

• Application of refinement techniques to a wider range of representational schema, viz. cases (REFINER++); constraints (CRIMSON); qualitative models (TIGON); and taxonomies (ReTAX).

• Production of a semi-formal framework (MUSKRAT) in which problem-solving, knowledge acquisition and knowledge refinement techniques can be represented.

• D Sleeman & F Mitchell, Towards Painless Knowledge Acquisition, in Advances in Knowledge Acquisition, N Shadbolt, K O'Hara & G Schreiber (Eds), Springer-Verlag, Berlin, 262-277, 1996.

• L Carbonara & D Sleeman, Stalker: An Efficient Knowledge Base Refinement System, in Proceedings of Workshop on Verification, Validation and Refinement of Knowledge Based Systems at Pacific Rim International Conference on AI, G Antoniou (Ed), 7-16, 1996.

• L Carbonara & D Sleeman, Improving the Efficiency of Knowledge Base Refinement, in Proceedings of the 13th International Conference on Machine Learning, L Saitta (Ed), Morgan Kaufmann, 78-86, 1996.

• P Leo, D Sleeman & A Tsinakos, S-SALT: A Problem-Solver, Knowledge Acquisition Tool and Associated Knowledge Base Refinement Mechanisms, in AI for Engineering Design, Analysis and Manufacturing, Cambridge University Press, 157-159, 1996.

• E Alberdi & D Sleeman, ReTAX: A Step in the Automation of Taxonomic Revision, Artificial Intelligence, 257-279, 1997.

• S Caggese, D Sleeman & F Mitchell, Workbench for Data Visualisation, Manipulation and Hypothesis Generation, in Proceedings of the Workshop on Automatic Learning and Natural Language, University of Torino, 141-144, 1997.

• D Sleeman & S White, Toolbox for Goal-Driven Knowledge Acquisition, in Proceedings of the 1997 Cognitive Science Society Conference, 1054, 1997.

• M Winter, D Sleeman & T Parsons, Inventory Management Using Constraint Satisfaction and Knowledge Refinement Techniques, Knowledge-Based Systems, 11, 293-300, 1998.

• S White & D Sleeman, Providing Advice on the Acquisition and Reuse of Knowledge Bases in Problem-Solving, in Proceedings of the 11th Banff Workshop on Knowledge Acquisition, Modelling and Management, B Gaines & M Musen (Eds), SRDG Publications, 1998.

• S White & D Sleeman, A Constraint-Based Approach to the Description of Competence, in Knowledge Acquisition, Modeling and Management, D Fensel & R Studer (Eds), 291-308, 1999.

• P Leo, S-SALT: A Problem Solver Plus Knowledge Acquisition Tool Which Additionally Can Refine its Knowledge Base, MSc, 1996.

• L Carbonara, Improving the Efficiency of Knowledge Base Refinement, PhD, 1997.

• F Mitchell, Painless Knowledge Acquisition for Time Series Data, PhD, 1998.

• S Metaxas, An Interactive Scientific Theory Revision System, PhD, 1998.

• M Winter, Knowledge Refinment in Constraint Satisfaction and Case Classification Problems, PhD, 1999.

The principle activities in this area have been:

• Development with a TMR fellow (V Corruble) of the FILTER system which helps a domain expert to retain or discard the many associations produced by data mining algorithms.

• Discovery via analogy and abstraction.

• The use of self-questioning as a means of performing focussed discovery.

• R Oehlmann, P Edwards & D Sleeman, Self-Questioning & Experimentation, in Progress in Case-Based Reasoning, I Watson (Ed), Springer-Verlag, 59-72, 1996.

• H A Simon, R E Valdez-Perez & D Sleeman, Scientific Discovery and Simplicity of Method (Guest Editorial), Artificial Intelligence, 91, 177-181, 1997.

• D Sleeman & V Corruble, The Role of Knowledge in a Data Mining Algorithm, in Proceedings of the 4th International Workshop on Multi-Strategy Learning (MSL-98), F Esposito, R Michalski & L Saitta (Eds), University of Torino, 165-174, 1998.

• A McQuatt, P J D Andrews, D Sleeman, V Corruble & P A Jones, The Analyses of Head Injury Data Using Decision Tree Techniques, in Proceedings of the AIMDM'99 Conference, W Horn et al (Eds), Springer-Verlag, 336-345, 1999.

• R Oehlmann, Exploratory Discovery: A Case-Based Planning Approach to Learning from Self- Questioning and Experimentation, PhD, 1996.

• D Roverso, Analogy by Mapping Spreading and Abstraction in Large, Multifunctional Knowledge Bases, PhD, 1997.

Nearest-neighbour/instance-based learning has been the focus of research into representational and dimensionality reduction issues. Specific topics:

• Set-based instance representation and distance metrics.

• Investigation of alternative methods for feature selection.

• Use of correspondence-analysis as a dimensionality reduction mechanism.

• T R Payne & P Edwards, Implicit Feature Selection with the Value Difference Metric, in Proceedings of European Conference on Artificial Intelligence, ECAI-98, 450-454, 1998.

• T R Payne & P Edwards, Dimensionality Reduction Through Sub-Space Mapping for Nearest Neighbour Algorithms, accepted for European Conference on Machine Learning (ECML-2000).

• T R Payne, Dimensionality Reduction & Representation for Nearest-Neighbour Learning, PhD, 1999.

The software agent paradigm has provided the context for a number of activities:

• Knowledge refinement and repair in a multi-agent context.

• Communication of inductive inferences between distributed agents.

• Induction of user interest profiles for intelligent information agents (I2As).

• Construction of a number of I2As with learning capabilities, including : LAW (Web assistant); ELVIS (travel advisor); Remora (personalised meta-search engine); COWBRA (collaborative Web page recommender).

• Integration of machine learning and information retrieval techniques.

• C L Green & P Edwards, Using Machine Learning to Enhance Software Tools for Internet Information Management, in AAAI-96 Workshop on Internet-Based Information Systems, A Franz & H Kitano (Eds), AAAI, 48-55, 1996.

• C Byrne & P Edwards, Refinement in Agent Groups, in Adaption & Learning in Multi-Agent Systems, G Weiss & S Sen (Eds), Lecture Notes in Artificial Intelligence 1042, Springer-Verlag, Berlin, 22-39, 1996.

• P Edwards, D Bayer, C L Green & T R Payne, Experience with Learning Agents which Manage Internet-Based Information, in AAAI 1996 Stanford Spring Symposium on Machine Learning in Information Access, M A Hearst & H Hirsh (Eds), AAAI, 31-40, 1996.

• P Edwards, C L Green, P C Lockier & T C Lukins, Exploiting Learning Technologies for World Wide Web Agents, in Proceedings of the IEE Colloquium on Intelligent World Wide Web Agents, Digest No: 97/118, IEE, Savoy Place, London, 3/1-3/7, 1997.

• W H E Davies & P Edwards, The Communication of Inductive Inferences, in Distributed Artificial Intelligence Meets Machine Learning: Learning in Multi-Agent Environments, G Weiss (Ed), Lecture Notes in Artificial Intelligence 1221, Springer-Verlag, Berlin, 223-241, 1997.

• T R Payne, P Edwards & C L Green, Experience with Rule Induction & k-Nearest Neighbour Methods for Interface Agents that Learn, IEEE Transactions on Knowledge & Data Engineering, 9 (2), 329-335, 1997.

• T R Payne & P Edwards, Interface Agents that Learn: An Investigation of Learning Issues in a Mail Agent Interface, Applied Artificial Intelligence, 11 (1), 1-32, 1997.

• W H E Davies & P Edwards, Communicating Inductive Inferences in Agent Societies, in Proceedings of the Agents'98 Workshop on Agents in Interaction - Acquiring Competence Through Imitation, K Dautenhahn & G Hayes (Eds), Minneapolis, USA, 1998.

• W H E Davies & P Edwards, Using Instance Selection To Combine Multiple Models Learned from Disjoint Subsets, in Instance Selection Methods, H Liu & H Motoda (Eds), Kluwer Scientific Publishers, expected: July 2000.

• P Edwards, M Steele, S E Clare & D M Johns, Exploiting a Web Cache for User Recommendations, submitted to Agents 2000 Workshop on Agent-Based Recommender Systems (WARS-2000).

Current active members of the Aberdeen group are listed below. Their Web pages contain additional information.