Information asymmetry in KL-regularized RL
This work addresses the challenge of inefficient learning in RL for tasks with repeated structure, offering a method to enhance training speed and performance, though it appears incremental as it builds on KL-regularized RL.
The paper tackles the problem of leveraging repeated structure in tasks to accelerate and regularize learning by learning a default policy with restricted information, which forces reusable behaviors. The result is a significant speedup and improvement in learning for certain tasks in both discrete and continuous action domains.
Many real world tasks exhibit rich structure that is repeated across different parts of the state space or in time. In this work we study the possibility of leveraging such repeated structure to speed up and regularize learning. We start from the KL regularized expected reward objective which introduces an additional component, a default policy. Instead of relying on a fixed default policy, we learn it from data. But crucially, we restrict the amount of information the default policy receives, forcing it to learn reusable behaviors that help the policy learn faster. We formalize this strategy and discuss connections to information bottleneck approaches and to the variational EM algorithm. We present empirical results in both discrete and continuous action domains and demonstrate that, for certain tasks, learning a default policy alongside the policy can significantly speed up and improve learning.