TY - GEN

T1 - Risk Bounds and Rademacher Complexity in Batch Reinforcement Learning

AU - Duan, Yaqi

AU - Jin, Chi

AU - Li, Zhiyuan

N1 - Funding Information:
ZL acknowledges support from NSF, ONR, Simons Foundation, Schmidt Foundation, Microsoft Research, Mozilla Research, Amazon Research, DARPA and SRC.
Publisher Copyright:
Copyright © 2021 by the author(s)

PY - 2021

Y1 - 2021

N2 - This paper considers batch Reinforcement Learning (RL) with general value function approximation. Our study investigates the minimal assumptions to reliably estimate/minimize Bellman error, and characterizes the generalization performance by (local) Rademacher complexities of general function classes, which makes initial steps in bridging the gap between statistical learning theory and batch RL. Concretely, we view the Bellman error as a surrogate loss for the optimality gap, and prove the followings: (1) In double sampling regime, the excess risk of Empirical Risk Minimizer (ERM) is bounded by the Rademacher complexity of the function class. (2) In the single sampling regime, sample-efficient risk minimization is not possible without further assumptions, regardless of algorithms. However, with completeness assumptions, the excess risk of FQI and a minimax style algorithm can be again bounded by the Rademacher complexity of the corresponding function classes. (3) Fast statistical rates can be achieved by using tools of local Rademacher complexity. Our analysis covers a wide range of function classes, including finite classes, linear spaces, kernel spaces, sparse linear features, etc.

AB - This paper considers batch Reinforcement Learning (RL) with general value function approximation. Our study investigates the minimal assumptions to reliably estimate/minimize Bellman error, and characterizes the generalization performance by (local) Rademacher complexities of general function classes, which makes initial steps in bridging the gap between statistical learning theory and batch RL. Concretely, we view the Bellman error as a surrogate loss for the optimality gap, and prove the followings: (1) In double sampling regime, the excess risk of Empirical Risk Minimizer (ERM) is bounded by the Rademacher complexity of the function class. (2) In the single sampling regime, sample-efficient risk minimization is not possible without further assumptions, regardless of algorithms. However, with completeness assumptions, the excess risk of FQI and a minimax style algorithm can be again bounded by the Rademacher complexity of the corresponding function classes. (3) Fast statistical rates can be achieved by using tools of local Rademacher complexity. Our analysis covers a wide range of function classes, including finite classes, linear spaces, kernel spaces, sparse linear features, etc.

UR - http://www.scopus.com/inward/record.url?scp=85161352529&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85161352529&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:85161352529

T3 - Proceedings of Machine Learning Research

SP - 2892

EP - 2902

BT - Proceedings of the 38th International Conference on Machine Learning, ICML 2021

PB - ML Research Press

T2 - 38th International Conference on Machine Learning, ICML 2021

Y2 - 18 July 2021 through 24 July 2021

ER -