Research

Vacancies and funding

I am no longer working in academic life. If you would like to continue working from this research, feel free to contact me and I will pass your name onto others who may be able to help.

Mainstream

Genetics-Based Machine Learning

My main research interest was in Genetics-based Machine Learning, with a principle focus on Learning Classifier Systems. Most of my work has examined the performance of one particular LCS implementation - XCS - in delayed-reward tasks. I have also applied XCS to the task of Data Mining. My last research interests in this area were:

• • Establishing a theoretical model for delayed reward learning in accuracy-based LCS

  • Identifying and evaluating methods for extending the length of path learning within XCS

  • Using principles from Grammatical Evolution to enhance the efficiency and effectiveness of Pittsburgh LCS approaches.

  • Hierarchy formation and exploitation in LCS

Biologically-Inspired Computing

A recent development, I was modelling the behaviour of large predators using A-Life techniques to identify the communications and social structures required to provide complex group behaviour with minimal cognition. The behaviours identified through simulation were implemented in robot simulators with the aim of application within real robots. My last research interests in this area were:

• • Applying identified predatory behaviours in land-based robots for searching and location

    • • see applet at http://www.cs.bath.ac.uk/~amb/lions/main.html

  • Demonstrating applications for identified behaviours within submersible robots

  • Expanding current findings in relation to other large land and marine predators

  • Linking with biologists and ethologists to investigate areas where the development of behavioural simulations has identified gaps in behavioural investigations

  • Using evolutionary computing to investigate the interaction between the development of social structures and the development of cooperative group behaviours.

Fading / Emerging

Data Mining

A fading interest latterly, but a useful application area for some of the techniques I looked at. I pioneered the use of accuracy-based LCS for Data Mining, and hope that my research to understand LCS better will lead to more reliable and effective GBML data mining techniques.

Agent-based Parallel Language

Going back to Minsky's ideas of Agent and Agency, and linking to computational-soup concepts, I developed and published a paper on a primitive highly parallel language. This language sought to create computationally simple "agents" with a very primitive idea of "intentionality" drawing on ideas from tuple-space programming and Hewitt's Actors language. It provided some neat ways of resolving deadlock and had potential as an algebra.

Past Funding

• • 2003 Royal Society Travel Grant - Travel grant for visit to conferences

  • Funding: £1,000

  • 2001 - 2001 UWE Studentship - Control Structure Emergence in Learning Classifier Systems

  • UWE Funding: £22,500

  • 1999 - 2001 Lead Teaching Company Scheme - Motorola MAST project. - Network Fault Diagnosis using Case Based Reasoning

  • Motorola/DTI Funding: £65,000

  • 1999 - 2001 PhD Student (Second supervisor) - Learning Classifier Systems for Mobile Robotics

  • (EPSRC/BT Funding [obtained by Dr L. Bull])

  • 1998 - 2000 Teaching Company Scheme - The Database Group - Data Mining using Genetic Algorithms.

  • DTI/SERC Funding : £65,000