Efficient Online-Bandit Strategies for Minimax Learning Problems

05/28/2021
āˆ™
by   Christophe Roux, et al.
āˆ™
0
āˆ™

Several learning problems involve solving min-max problems, e.g., empirical distributional robust learning or learning with non-standard aggregated losses. More specifically, these problems are convex-linear problems where the minimization is carried out over the model parameters wāˆˆš’² and the maximization over the empirical distribution pāˆˆš’¦ of the training set indexes, where š’¦ is the simplex or a subset of it. To design efficient methods, we let an online learning algorithm play against a (combinatorial) bandit algorithm. We argue that the efficiency of such approaches critically depends on the structure of š’¦ and propose two properties of š’¦ that facilitate designing efficient algorithms. We focus on a specific family of sets š’®_n,k encompassing various learning applications and provide high-probability convergence guarantees to the minimax values.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset