Convex Repeated Games and Fenchel Duality
Shai Shalev-Shwartz and Yoram Singer
In: NIPS 2006, 4-10 Dec 2006, Vancouver, CA.

Abstract

We describe and analyze an algorithmic framework for playing convex repeated games. In each trial of the repeated game, the first player predicts a vector and then the second player responds with a loss function over the vector. Based on a generalization of Fenchel duality, we derive an algorithmic framework for the first player and analyze the player's regret. We then use our algorithmic framework and its corresponding regret analysis for online learning problems and for boosting.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.

[Edit]