DATA SENSITIVE ANALYSIS OF GENERALIZATION
Professor John Shawe-Taylor, Dept of Computer Science, Royal Holloway, University of London

Abstract:Until recently the distribution free or probably approximately correct model of learning was only able to give a priori bounds on generalization based on the complexity of the space of hypotheses, measured either by its size or its Vapnik-Chervonenkis dimension. New results will be presented which allow better bounds for the generalization performance to be obtained in terms of properties of the trained system, something that was previously only possible using Bayesian analysis. The talk will describe the approach and discuss its applications to model selection as well as confidence estimation of the classification of particular inputs.

Details: Wednesday 26 February, 2.30 pm, Seminar Room, Bedford Library, Royal Holloway, University of London

back