Back to search

Georgia Institute of Technology

Risk-Averse Bayesian Optimal Control under Distributional Uncertainty

Abstract

dc:description.abstract

We study infinite-horizon stochastic optimal control problems under distributional uncertainty, where system randomness is governed by unknown model parameters that are learned sequentially from data. With limited data, substantial epistemic uncertainty may remain, and adverse parameter realizations can lead to significantly higher long-run costs. To address this issue, we develop a risk-averse episodic Bayesian optimal control framework in which the posterior distribution of the unknown parameters is updated across episodes, and decisions are computed through a Bellman equation that applies Conditional Value-at-Risk (CVaR) with respect to the posterior distribution of the unknown parameters. Theoretically, we establish an asymptotic convergence rate of order OP\star\otimes N(N-1/2) for the Bayesian risk-averse value function and derive an asymptotic distribution result for the associated error term in policy evaluation. Under exponential-family models, we further derive finite-time regret bounds of order \widetilde O\left(\frac{\sqrt{N}}{(1-α)(1-γ)}\right), clarifying how learning, discounting, and risk aversion jointly affect performance. Computationally, we adapt a stochastic dual dynamic programming (SDDP) method to solve the resulting risk-averse Bayesian control problem in the convex setting and develop a cut-reuse mechanism to improve sample efficiency across posterior updates. Numerical experiments on a risk-averse inventory control problem verify the base-stock structure and support the theoretical results.

Degree

thesis:*
Name thesis:degree_name
Mathematics, MS
Department thesis:degree_department
Mathematics
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Deng, Chuyun
Advisors dc:contributor.advisor
  • Zhou, Enlu
  • Zhou, Haomin
Committee members dc:contributor.committeemember
  • Shapiro, Alexander
  • Liao, Wenjing
ORCID person:identifier.orcid

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/81595

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Deng, Chuyun. Risk-Averse Bayesian Optimal Control under Distributional Uncertainty. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/81595