Agnostic Q-learning with Function Approximation in Deterministic Systems: Near-Optimal Bounds on Approximation Error and Sample Complexity

Abstract

The current paper studies the problem of agnostic Q-learning with function approximation in deterministic systems where the optimal Q-function is approximable by a function in the class F with approximation error 0. We propose a novel recursion-based algorithm and show that if = O(/_E), then one can find the optimal policy using O(_E) trajectories, where is the gap between the optimal Q-value of the best actions and that of the second-best actions and _E is the Eluder dimension of F. Our result has two implications: enumerate In conjunction with the lower bound in [Du et al., 2020], our upper bound suggests that the condition = (/_E) is necessary and sufficient for algorithms with polynomial sample complexity. In conjunction with the obvious lower bound in the tabular case, our upper bound suggests that the sample complexity (_E) is tight in the agnostic setting. enumerate Therefore, we help address the open problem on agnostic Q-learning proposed in [Wen and Van Roy, 2013]. We further extend our algorithm to the stochastic reward setting and obtain similar results.

Tasks

Reproductions