Uniform Last-Iterate Guarantee for Bandits and Reinforcement Learning

This paper introduces a stronger metric, uniform last-iterate (ULI) guarantee, capturing both cumulative and instantaneous performance of RL algorithms.