We introduce confidence-weighted linear classifiers, a new class of algorithms that maintain confidence information about classifier parameters. Learning in this framework updates parameters by estima
