Speaker Sebastien Bubeck
Host Ofer Dekel
Affiliation Princeton University
Date recorded 21 January 2014
The linear bandit problem is a far-reaching extension of the classical multi-armed bandit problem. In the recent years linear bandits have emerged as a core problem of sequential decision making, somewhat analogously to what happened with linear programming in optimization or linear regression in statistics. Despite its importance we still do not have a complete picture for this problem: in some cases we have optimal strategies (from an information theoretic point of view) but they are algorithmically intractable, while in other cases we even lack information optimal strategies. In this talk I will describe precisely the situation where we stand and the contributions I made to this problem.
©2014 Microsoft Corporation. All rights reserved.