New math explains why Q-learning overestimates value

Original: Sign-Separated Finite-Time Error Analysis of Q-Learning

Source: arxiv.org ↗