progressive point matching

Preston Fu proposed progressive point matching (PPM), an asymptotically unbiased framework for assigning partial credit in reinforcement learning of long-horizon language model tasks, published in Sep…