Minimax Policies for Combinatorial Prediction Games

Jean-Yves Audibert 1, 2 Sébastien Bubeck 3 Gábor Lugosi 4, 5
1 IMAGINE [Marne-la-Vallée]
LIGM - Laboratoire d'Informatique Gaspard-Monge, CSTB - Centre Scientifique et Technique du Bâtiment, ENPC - École des Ponts ParisTech
2 SIERRA - Statistical Machine Learning and Parsimony
DI-ENS - Département d'informatique de l'École normale supérieure, ENS Paris - École normale supérieure - Paris, Inria Paris-Rocquencourt, CNRS - Centre National de la Recherche Scientifique : UMR8548
Abstract : We address the online linear optimization problem when the actions of the forecaster are represented by binary vectors. Our goal is to understand the magnitude of the minimax regret for the worst possible set of actions. We study the problem under three different assumptions for the feedback: full information, and the partial information models of the so-called "semi-bandit", and "bandit" problems. We consider both L -, and L2-type of restrictions for the losses assigned by the adversary. We formulate a general strategy using Bregman projections on top of a potential-based gradient descent, which generalizes the ones studied in the series of papers Gyorgy et al. (2007), Dani et al. (2008), Abernethy et al. (2008), Cesa-Bianchi and Lugosi (2009), Helmbold and Warmuth (2009), Koolen et al. (2010), Uchiya et al. (2010), Kale et al. (2010) and Audibert and Bubeck (2010). We provide simple proofs that recover most of the previous results. We propose new upper bounds for the semi-bandit game. Moreover we derive lower bounds for all three feedback assumptions. With the only exception of the bandit game, the upper and lower bounds are tight, up to a constant factor. Finally, we answer a question asked by Koolen et al. (2010) by showing that the exponentially weighted average forecaster is suboptimal against L adversaries.
Type de document :
Communication dans un congrès
COLT - 24th Conference on Learning Theory - 2011, Jul 2011, Budapest, Hungary. in press, 2011
Liste complète des métadonnées
Contributeur : Jean-Yves Audibert <>
Soumis le : samedi 17 septembre 2011 - 22:59:32
Dernière modification le : lundi 28 janvier 2019 - 09:03:40


  • HAL Id : hal-00624463, version 1


Jean-Yves Audibert, Sébastien Bubeck, Gábor Lugosi. Minimax Policies for Combinatorial Prediction Games. COLT - 24th Conference on Learning Theory - 2011, Jul 2011, Budapest, Hungary. in press, 2011. 〈hal-00624463〉



Consultations de la notice