Bias-variance decompositions: the exclusive privilege of Bregman divergences

Tom Heskes

Bias-variance decompositions are widely used to understand the generalization performance of machine learning models. While the squared error loss permits a straightforward decompo…