Off-Policy Actor-Critic with Shared Experience Replay
This work addresses stability and efficiency issues in reinforcement learning for researchers and practitioners, though it appears incremental as it builds on existing methods like V-trace.
The paper tackles the challenges of combining actor-critic reinforcement learning with large-scale experience replay, proposing solutions for efficient learning and stability in off-policy settings, and demonstrates state-of-the-art data efficiency on Atari with training up to 200M frames.
We investigate the combination of actor-critic reinforcement learning algorithms with uniform large-scale experience replay and propose solutions for two challenges: (a) efficient actor-critic learning with experience replay (b) stability of off-policy learning where agents learn from other agents behaviour. We employ those insights to accelerate hyper-parameter sweeps in which all participating agents run concurrently and share their experience via a common replay module. To this end we analyze the bias-variance tradeoffs in V-trace, a form of importance sampling for actor-critic methods. Based on our analysis, we then argue for mixing experience sampled from replay with on-policy experience, and propose a new trust region scheme that scales effectively to data distributions where V-trace becomes unstable. We provide extensive empirical validation of the proposed solution. We further show the benefits of this setup by demonstrating state-of-the-art data efficiency on Atari among agents trained up until 200M environment frames.