This paper delves into the partial monitoring (PM) framework and its application to sequential learning problems with incomplete feedback. The authors introduce a new class of strategies based on the randomization of deterministic confidence bounds, extending regret guarantees. The study shows that the proposed RandCBP and RandCBPside strategies improve current baselines in PM games. The authors also establish a use case for the PM framework in monitoring the error rate of deployed classification systems.
Publication date: 8 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.05002
