Landscape Analysis of Bayesian Network Structure Learning Algorithms
- When?
- Wednesday 22 June 2011, 13:30 to 14:30
- Open to:
- Staff, Students
- Speaker:
- Professor John McCall, Robert Gordon University
Bayesian Networks (BN) are an increasingly important tool for mining complex relations in large data sets. A major focus of current research is in efficient and effective ways of learning those essential interactions between variables, known as structure, that allow efficient factorisation of the joint probability distribution of the data. This in turn provides a platform for prediction, inference and simulation.
Search-and-score algorithms using nature-inspired metaheuristics are a leading technique for BN structure learning, however performance is variable and strongly problem-dependent. In this talk I will show how fitness landscape analysis can be used to explain empirically-observed performance differences between particular search-and-score algorithms on some well-studied benchmark problems. We investigate the average landscape discovered by random walks around optimal points in the space of BN node orderings. Differences in algorithm performance are explained in terms of these landscapes, which in turn are related to properties of the BN structures. These initial findings suggest that fitness landscape analysis is a promising approach for explaining existing empirical performance comparisons with further potential for adaptive search in problems with unknown structure.
John McCall is a Professor of Computing at Robert Gordon University where he leads research in Digital Technologies in the IDEAS Research Institute. Originally a pure mathematician, his main focus of research for over 15 years has been in nature-inspired computing, including evolutionary algorithms, particle swarms, ant colonies and estimation of distribution algorithms (EDA). His particular interests include EA simulation wrapper approaches, the fusion between EDA and probabilistic modelling, Markov Network EDAs and data modelling. Prof. McCall is involved with a wide range of EA applications to industry. He pioneered the application of EA to problems in optimal cancer chemotherapy control and is actively involved in a number of medical treatment optimisation projects. Other application areas include oil and gas decision support and scheduling. Current collaborators include British Telecom, ODS-Petrodata, Viper Technologies, NHS and the British Association of Urological Surgeons.