Cost-sensitive learning based on Bregman divergences

This paper analyzes the application of a particular class of Bregman divergences to design cost-sensitive classifiers for multiclass problems. We show that these divergence measures can be used to est