Method separates a policy's real skill from lucky timing

Original: Regret-based Characterization, Information Limits, and the Sample Complexity of Auditing Adaptive Policies

Source: arxiv.org ↗