Finite-Time Error Bounds for Greedy-GQ

6 Sep 2022  ·  Yue Wang, Yi Zhou, Shaofeng Zou ·

Greedy-GQ with linear function approximation, originally proposed in \cite{maei2010toward}, is a value-based off-policy algorithm for optimal control in reinforcement learning, and it has a non-linear two timescale structure with a non-convex objective function. This paper develops its finite-time error bounds. We show that the Greedy-GQ algorithm converges as fast as $\mathcal{O}({1}/{\sqrt{T}})$ under the i.i.d.\ setting and $\mathcal{O}({\log T}/{\sqrt{T}})$ under the Markovian setting. We further design a variant of the vanilla Greedy-GQ algorithm using the nested-loop approach, and show that its sample complexity is $\mathcal{O}({\log(1/\epsilon)\epsilon^{-2}})$, which matches with the one of the vanilla Greedy-GQ. Our finite-time error bounds match with one of the stochastic gradient descent algorithms for general smooth non-convex optimization problems. Our finite-sample analysis provides theoretical guidance on choosing step-sizes for faster convergence in practice and suggests the trade-off between the convergence rate and the quality of the obtained policy. Our techniques in this paper provide a general approach for finite-sample analysis of non-convex two timescale value-based reinforcement learning algorithms.

PDF Abstract

Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods


No methods listed for this paper. Add relevant methods here