expectation-consistency-loss

This paper introduces the Expectation Consistency Loss (ECL), a theoretically grounded loss function for calibrating classifier confidence under covariate shift, derived from a necessary and sufficient condition called the Expectation Consistency Condition.