Dr Gavin Brown University of Manchester School of Computer Science Kilburn Building Oxford Road Manchester M13 9PL 0161 275 6190
Publications    Projects + Grants    Research Students    Collaborators I am a member of the Machine Learning and Optimization Group. My research interests can be summarised as: feature selection/extraction with information theoretic methods, Markov Blanket algorithms, ensemble learning (aka multiple classifier systems), and online learning. All of the above in application to two domains: Systems Biology and adaptive compiler optimisation. Recent Research Activities: New book chapter - Ensemble Learning I wrote an invited chapter for the forthcoming Springer Encyclopedia of Machine Learning. "The study of ensemble methods, with model outputs considered for their abstract properties rather than the specifics of the algorithm which produced them, allows for a wide impact across many fields of study. If we can understand precisely why, when, and how particular ensemble methods can be applied successfully, we would have made progress toward a powerful new tool for Machine Learning: the ability to automatically exploit the strengths and weaknesses of different learning systems." AISTATS 2009 paper - Feature Selection with Information Theory The traditional approach to so-called filter methods in feature selection is to construct a criterion to measure the utility of any given feature. The more sophisticated methods penalize feature-feature correlations (`redundancy') with various penalty terms. The last 15 years have produced a flood of papers advocating different penalty terms. My recent work shows that the vast majority of these can be naturally derived from a single framework, using multivariate information theory. The work reveals that there exists a natural, smooth space space of feature selection criteria, where each paper over the last 15 years corresponds to one point. Most of the space has never been explored. See the AISTATS 2009 paper for details. Invited plenary at UK-KDD 2009 - Feature Selection by Filters, a Unifying Perspective I gave an invited talk at UK-KDD 2009. New Grant - Dynamic Ensemble Techniques (EPSRC grant EP/F023855/1) With colleagues at Bristol, I am investigating how dynamic ensemble techniques can tackle multi-step (control) and nonstationary problems. This is in collaboration with Tim Kovacs, James Marshall and Jeremy Wyatt, conducted under our EPSRC funded ADEPT project. New Grant - Machine Learning for Multi-Core Computers (EPSRC grant EP/G000662/1) The computer industry is undergoing the "multi-core" revolution. When you buy a PC off the shelf these days, it is inevitably "dual-core" or "quad-core". This idea of more and more CPU "cores" executing in parallel is expected to continue to the hundreds and thousands. The problem of coordinating these cores is challenging and unsolved. With Mikel Lujan and Jeremy Singer I am working on applying Machine Learning to this problem, conducted under our EPSRC funded iTLS project. IEEE TNN paper on Sparse Distributed Memories In a project with Steve Furber I found that sparse distributed memory models like the correlation matrix memories of Wilshaw and Kanerva could give significant insights into the design of fault tolerant computer architectures. This resulted in a IEEE TNN paper available here. Ensemble Learning I worked for a long while on the issue of diversity in ensembles, with Jeremy Wyatt. A summary of the work can be found on this page. A slightly less optimistic (but rather insightful) take on the field is found here. Image Feature Extraction I did a nice project with Honda several years ago, which turned into a patent, on image feature extraction - I follow up little avenues on this occasionally. Throughout this time I have maintained an interest in evolutionary speciation and optimisation, which has spun off into several useful collaborations.