Options
Minimax-Bayes Reinforcement Learning
Auteur(s)
Thomas Kleine Buening
Hannes Eriksson
Divya Grover
Emilio Jorge
Maison d'édition
PMLR
Date de parution
2023
In
Proceedings of Machine Learning Research
Vol.
206
De la page
7511
A la page
7527
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.
Notes
International Conference on Artificial Intelligence and Statistics, 25-27 April 2023, Palau de Congressos, Valencia, Spain
Identifiants
Autre version
https://proceedings.mlr.press/v206/buening23a.html
Type de publication
conference paper
Dossier(s) à télécharger
