Leverage the Average: an Analysis of KL Regularization in RL
This provides foundational insights for RL practitioners by explaining why KL regularization improves performance, though it is incremental as it builds on existing methods.
The paper tackles the theoretical understanding of KL regularization in reinforcement learning by showing it implicitly averages q-values, resulting in a performance bound with linear horizon dependency and error averaging instead of accumulation.
Recent Reinforcement Learning (RL) algorithms making use of Kullback-Leibler (KL) regularization as a core component have shown outstanding performance. Yet, only little is understood theoretically about why KL regularization helps, so far. We study KL regularization within an approximate value iteration scheme and show that it implicitly averages q-values. Leveraging this insight, we provide a very strong performance bound, the very first to combine two desirable aspects: a linear dependency to the horizon (instead of quadratic) and an error propagation term involving an averaging effect of the estimation errors (instead of an accumulation effect). We also study the more general case of an additional entropy regularizer. The resulting abstract scheme encompasses many existing RL algorithms. Some of our assumptions do not hold with neural networks, so we complement this theoretical analysis with an extensive empirical study.