Maximin Action Identification: A New Bandit Framework for Games - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

Maximin Action Identification: A New Bandit Framework for Games

Résumé

We study an original problem of pure exploration in a strategic bandit model motivated by Monte Carlo Tree Search. It consists in identifying the best action in a game, when the player may sample random outcomes of sequentially chosen pairs of actions. We propose two strategies for the fixed-confidence setting: Maximin-LUCB, based on lower-and upper-confidence bounds; and Maximin-Racing, which operates by successively eliminating the sub-optimal actions. We discuss the sample complexity of both methods and compare their performance empirically. We sketch a lower bound analysis, and possible connections to an optimal algorithm.
Fichier principal
Vignette du fichier
garivier16b.pdf (253.94 Ko) Télécharger le fichier
Origine Fichiers éditeurs autorisés sur une archive ouverte
Loading...

Dates et versions

hal-01273842 , version 1 (14-02-2016)
hal-01273842 , version 2 (21-11-2016)

Identifiants

Citer

Aurélien Garivier, Emilie Kaufmann, Wouter M. Koolen. Maximin Action Identification: A New Bandit Framework for Games. 29th Annual Conference on Learning Theory (COLT), Jun 2016, New-York, United States. ⟨hal-01273842v2⟩
1665 Consultations
182 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More