Options
Minimax-Bayes Reinforcement Learning
Auteur(s)
Thomas Kleine Buening
University of Oslo
Hannes Eriksson
Zenseact
Divya Grover
Emilio Jorge
Chalmers University of Technology
Date de parution
2023
In
PMLR
Vol.
206
Résumé
While the Bayesian decision-theoretic framework offers an elegant solution to the problem of decision making under uncertainty, one question is how to appropriately select the prior distribution. One idea is to employ a worst-case prior. However, this is not as easy to specify in sequential decision making as in simple statistical estimation problems. This paper studies (sometimes approximate) minimax-Bayes solutions for various reinforcement learning problems to gain insights into the properties of the corresponding priors and policies. We find that while the worst-case prior depends on the setting, the corresponding minimax policies are more robust than those that assume a standard (i.e. uniform) prior.
Nom de l'événement
AISTATS
Lieu
Valencia, Spain
Identifiants
Type de publication
conference proceedings
Dossier(s) à télécharger
