Predictor-Rejector Multi-Class Abstention: Theoretical Analysis and Algorithms

arXiv:2310.14772v2 Announce Type: replace
Abstract: We study the key framework of learning with abstention in the multi-class classification setting. In this setting, the learner can choose to abstain from making a prediction with some pre-defined cost. We present a series of new theoretical and algorithmic results for this learning problem in the predictor-rejector framework. We introduce several new families of surrogate losses for which we prove strong non-asymptotic and hypothesis set-specific consistency guarantees, thereby resolving positively two existing open questions. …

abstract algorithms analysis arxiv class classification cost cs.lg framework key making prediction results series stat.ml study the key type