Revisiting the Softmax Bellman Operator: New Benefits and New Perspective
[edit]
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:59165925, 2019.
Abstract
The impact of softmax on the value function itself in reinforcement learning (RL) is often viewed as problematic because it leads to suboptimal value (or Q) functions and interferes with the contraction properties of the Bellman operator. Surprisingly, despite these concerns, and independent of its effect on exploration, the softmax Bellman operator when combined with Deep Qlearning, leads to Qfunctions with superior policies in practice, even outperforming its double Qlearning counterpart. To better understand how and why this occurs, we revisit theoretical properties of the softmax Bellman operator, and prove that (i) it converges to the standard Bellman operator exponentially fast in the inverse temperature parameter, and (ii) the distance of its Q function from the optimal one can be bounded. These alone do not explain its superior performance, so we also show that the softmax operator can reduce the overestimation error, which may give some insight into why a suboptimal operator leads to better performance in the presence of value function approximation. A comparison among different Bellman operators is then presented, showing the tradeoffs when selecting them.
Related Material


