Faculty of Engineering and Natural Sciences
Ensemble MLP Classifier Design
The idea of combining multiple classifiers is based on the observation that achieving optimal performance in combination is not necessarily consistent with obtaining the best performance for a single (base) classifier. However, the base classifier parameters still need to be set, and the optimal parameters may be different for the ensemble. The normal way to set parameters is to use a validation set or cross-validation techniques. In this talk, measures for setting parameters for two-class problems will be discussed, and extended to the problem of identifying and removing irrelevant features. The technique is extended to multi-class problems using ECOC (Error-Correcting Output Coding). Examples using MLP base classifiers for face recognition will be described.
Terry Windeatt received the B.Sc. degree in Applied Science from University of Sussex, followed by M.Sc. in Electronic Engineering from University of California and a PhD from University of Surrey, U.K. After lecturing at Kingston University, UK, he worked in the USA for eight years, at the Research and Development Departments of General Motors and Xerox Corporation in Rochester, NY. His industrial R&D experience is in modeling/simulation for intelligent automotive and office-copying applications. He returned to the UK in 1984 to the University of Surrey, where he now lectures in Machine Intelligence. He has worked on various research projects at the CVSSP, and his current research interests include neural nets, pattern recognition, and computer vision, with emphasis to facial expression analysis.
June 3, 2008, 10:40, FENS G032