Bounding the Capacity Measure of Multi-Class Discriminant Models
Abstract
Vapnik's statistical learning theory has mainly been developed for two types of problems: pattern recognition (computation of dichotomies) and regression (estimation of real-valued functions). Multi-class discriminant analysis has only been studied independently in recent years. Extending several standard results, among which a famous theorem by Bartlett, we have derived distribution-free uniform strong laws of large numbers devoted to multi-class discriminant models. This technical report deals with the computation of the capacity measures involved in these bounds on the expected risk. It considers more specifically the case of multi-class SVMs.