Recursive Entropic Risk Optimization in Discounted MDPs: Sample Complexity Bounds with a Generative Model

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Mortensen, Oliver, Talebi, Mohammad Sadegh
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911695076589568
author Mortensen, Oliver
Talebi, Mohammad Sadegh
author_facet Mortensen, Oliver
Talebi, Mohammad Sadegh
contents We study risk-sensitive reinforcement learning in finite discounted MDPs with recursive entropic risk measures (ERM), where the risk parameter $β\neq 0$ controls the agent's risk attitude: $β>0$ for risk-averse and $β<0$ for risk-seeking behavior. A generative model of the MDP is assumed to be available. Our focus is on the sample complexities of learning the optimal state-action value function (value learning) and an optimal policy (policy learning) under recursive ERM. We introduce a model-based algorithm, called Model-Based ERM $Q$-Value Iteration (MB-RS-QVI), and derive PAC-type bounds on its sample complexity for both value and policy learning. Both PAC bounds scale exponentially with $|β|/(1-γ)$, where $γ$ is the discount factor. We also establish corresponding lower bounds for both value and policy learning, showing that exponential dependence on $|β|/(1-γ)$ is unavoidable in the worst case. The bounds are tight in the number of states and actions ($S$ and $A$), providing the first rigorous sample complexity guarantees for recursive ERM across both risk-averse and risk-seeking regimes.
format Preprint
id arxiv_https___arxiv_org_abs_2506_00286
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Recursive Entropic Risk Optimization in Discounted MDPs: Sample Complexity Bounds with a Generative Model
Mortensen, Oliver
Talebi, Mohammad Sadegh
Machine Learning
Artificial Intelligence
Optimization and Control
We study risk-sensitive reinforcement learning in finite discounted MDPs with recursive entropic risk measures (ERM), where the risk parameter $β\neq 0$ controls the agent's risk attitude: $β>0$ for risk-averse and $β<0$ for risk-seeking behavior. A generative model of the MDP is assumed to be available. Our focus is on the sample complexities of learning the optimal state-action value function (value learning) and an optimal policy (policy learning) under recursive ERM. We introduce a model-based algorithm, called Model-Based ERM $Q$-Value Iteration (MB-RS-QVI), and derive PAC-type bounds on its sample complexity for both value and policy learning. Both PAC bounds scale exponentially with $|β|/(1-γ)$, where $γ$ is the discount factor. We also establish corresponding lower bounds for both value and policy learning, showing that exponential dependence on $|β|/(1-γ)$ is unavoidable in the worst case. The bounds are tight in the number of states and actions ($S$ and $A$), providing the first rigorous sample complexity guarantees for recursive ERM across both risk-averse and risk-seeking regimes.
title Recursive Entropic Risk Optimization in Discounted MDPs: Sample Complexity Bounds with a Generative Model
topic Machine Learning
Artificial Intelligence
Optimization and Control
url https://arxiv.org/abs/2506.00286