W. Dabney, “The Value-Improvement Path: Towards Better Representations for Reinforcement Learning”, AAAI, vol. 35, no. 8, pp. 7160-7168, May 2021.