TU Darmstadt / ULB / TUprints

f-Divergence constrained policy improvement

Belousov, Boris ; Peters, Jan (2023)
f-Divergence constrained policy improvement.
doi: 10.26083/tuprints-00020553
Report, Secondary publication, Preprint

[img] Text
1801.00056.pdf
Copyright Information: CC BY 4.0 International - Creative Commons, Attribution.

Download (693kB)
Item Type: Report
Type of entry: Secondary publication
Title: f-Divergence constrained policy improvement
Language: English
Date: 17 October 2023
Place of Publication: Darmstadt
Year of primary publication: 2017
Collation: 20 Seiten
DOI: 10.26083/tuprints-00020553
Corresponding Links:
Origin: Secondary publication service
Abstract:

To ensure stability of learning, state-of-the-art generalized policy iteration algorithms augment the policy improvement step with a trust region constraint bounding the information loss. The size of the trust region is commonly determined by the Kullback-Leibler (KL) divergence, which not only captures the notion of distance well but also yields closed-form solutions. In this paper, we consider a more general class of f-divergences and derive the corresponding policy update rules. The generic solution is expressed through the derivative of the convex conjugate function to f and includes the KL solution as a special case. Within the class of f-divergences, we further focus on a one-parameter family of α-divergences to study effects of the choice of divergence on policy improvement. Previously known as well as new policy updates emerge for different values of α. We show that every type of policy update comes with a compatible policy evaluation resulting from the chosen f-divergence. Interestingly, the mean-squared Bellman error minimization is closely related to policy evaluation with the Pearson χ²-divergence penalty, while the KL divergence results in the soft-max policy update and a log-sum-exp critic. We carry out asymptotic analysis of the solutions for different values of α and demonstrate the effects of using different divergence functions on a multi-armed bandit problem and on common standard reinforcement learning problems.

Uncontrolled Keywords: Reinforcement Learning, Policy Search, Bandit Problems
Status: Preprint
URN: urn:nbn:de:tuda-tuprints-205534
Classification DDC: 000 Generalities, computers, information > 004 Computer science
Divisions: 20 Department of Computer Science > Intelligent Autonomous Systems
TU-Projects: EC/H2020|640554|SKILLS4ROBOTS
Date Deposited: 17 Oct 2023 15:10
Last Modified: 23 Oct 2023 09:21
URI: https://tuprints.ulb.tu-darmstadt.de/id/eprint/20553
PPN: 512614172
Export:
Actions (login required)
View Item View Item