Functional Acceleration for Policy Mirror Descent
This work addresses the challenge of improving efficiency in reinforcement learning algorithms, particularly for large-scale optimization, though it appears incremental as it builds on existing PMD methods.
The paper tackled the problem of accelerating policy optimization in reinforcement learning by applying functional acceleration to the Policy Mirror Descent family, resulting in a momentum-based update that is independent of policy parametrization and applicable to large-scale settings, with theoretical analysis and numerical studies illustrating its dynamics.
We apply functional acceleration to the Policy Mirror Descent (PMD) general family of algorithms, which cover a wide range of novel and fundamental methods in Reinforcement Learning (RL). Leveraging duality, we propose a momentum-based PMD update. By taking the functional route, our approach is independent of the policy parametrization and applicable to large-scale optimization, covering previous applications of momentum at the level of policy parameters as a special case. We theoretically analyze several properties of this approach and complement with a numerical ablation study, which serves to illustrate the policy optimization dynamics on the value polytope, relative to different algorithmic design choices in this space. We further characterize numerically several features of the problem setting relevant for functional acceleration, and lastly, we investigate the impact of approximation on their learning mechanics.