Beyond Average Return in Markov Decision Processes - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Beyond Average Return in Markov Decision Processes

Résumé

What are the functionals of the reward that can be computed and optimized exactly in Markov Decision Processes? In the finite-horizon, undiscounted setting, Dynamic Programming (DP) can only handle these operations efficiently for certain classes of statistics. We summarize the characterization of these classes for policy evaluation, and give a new answer for the planning problem. Interestingly, we prove that only generalized means can be optimized exactly, even in the more general framework of Distributional Reinforcement Learning (DistRL). DistRL permits, however, to evaluate other functionals approximately. We provide error bounds on the resulting estimators, and discuss the potential of this approach as well as its limitations. These results contribute to advancing the theory of Markov Decision Processes by examining overall characteristics of the return, and particularly risk-conscious strategies.
Fichier principal
Vignette du fichier
main_neurips.pdf (517.58 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04264487 , version 1 (30-10-2023)
hal-04264487 , version 2 (16-02-2024)

Licence

Paternité - Partage selon les Conditions Initiales

Identifiants

Citer

Alexandre Marthe, Aurélien Garivier, Claire Vernade. Beyond Average Return in Markov Decision Processes. Neurips 2023, Dec 2023, New Orleans, United States. ⟨hal-04264487v2⟩
24 Consultations
56 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More