Skip to content

Latest commit

 

History

History
9 lines (6 loc) · 706 Bytes

COF-PAC.md

File metadata and controls

9 lines (6 loc) · 706 Bytes

Provably Convergent Off-Policy Actor-Critic with Function Approximation

Shangtong Zhang, Bo Liu, Hengshuai Yao, Shimon Whiteson

Abstract

We present the first provably convergent offpolicy actor-critic algorithm (COF-PAC) with function approximation in a two-timescale form.

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.