Gradient Temporal-Difference Learning with Regularized Corrections
This addresses the dilemma for practitioners in reinforcement learning who need stable, easy-to-use algorithms, though it is incremental as it builds on existing gradient TD methods.
The authors tackled the instability of Q-learning and TD learning in large neural networks by introducing TD with Regularized Corrections (TDRC), which matches TD's performance when stable but prevents divergence in unstable cases, showing gradient TD methods can be a better alternative.
It is still common to use Q-learning and temporal difference (TD) learning-even though they have divergence issues and sound Gradient TD alternatives exist-because divergence seems rare and they typically perform well. However, recent work with large neural network learning systems reveals that instability is more common than previously thought. Practitioners face a difficult dilemma: choose an easy to use and performant TD method, or a more complex algorithm that is more sound but harder to tune and all but unexplored with non-linear function approximation or control. In this paper, we introduce a new method called TD with Regularized Corrections (TDRC), that attempts to balance ease of use, soundness, and performance. It behaves as well as TD, when TD performs well, but is sound in cases where TD diverges. We empirically investigate TDRC across a range of problems, for both prediction and control, and for both linear and non-linear function approximation, and show, potentially for the first time, that gradient TD methods could be a better alternative to TD and Q-learning.