English

Actor-Critic Algorithm for Dynamic Expectile and CVaR

Machine Learning 2026-05-11 v1

Abstract

Optimizing dynamic risk with stochastic policies is challenging in both policy updates and value learning. The former typically requires transition perturbation, while the latter may rely on model-based approaches. To address these challenges, we propose a surrogate policy gradient without transition perturbation under softmax policy parameterization. We further develop model-free value learning methods for dynamic expectile and conditional value-at-risk by leveraging elicitability. Finally, inspired by Expected SARSA and Expected Policy Gradient, a model-free off-policy actor-critic algorithm is constructed. Empirical results in domains with verifiable risk-averse behavior show that our algorithm can learn risk-averse policy and consistently outperforms other existing methods.

Keywords

Cite

@article{arxiv.2605.07857,
  title  = {Actor-Critic Algorithm for Dynamic Expectile and CVaR},
  author = {Yudong Luo and Erick Delage},
  journal= {arXiv preprint arXiv:2605.07857},
  year   = {2026}
}
R2 v1 2026-07-01T12:57:57.821Z