Provably Convergent Two-Timescale Off-Policy Actor-Critic with Function Approximation
This provides a theoretical foundation for off-policy reinforcement learning, addressing a key bottleneck in the field, though it appears incremental as it builds on existing methods like Gradient TD and Emphatic TD.
The paper tackles the challenge of developing a provably convergent off-policy actor-critic algorithm with function approximation, resulting in COF-PAC, which introduces an emphasis critic trained via Gradient Emphasis Learning and demonstrates convergence with linear critics and a nonlinear actor.
We present the first provably convergent two-timescale off-policy actor-critic algorithm (COF-PAC) with function approximation. Key to COF-PAC is the introduction of a new critic, the emphasis critic, which is trained via Gradient Emphasis Learning (GEM), a novel combination of the key ideas of Gradient Temporal Difference Learning and Emphatic Temporal Difference Learning. With the help of the emphasis critic and the canonical value function critic, we show convergence for COF-PAC, where the critics are linear and the actor can be nonlinear.