Combinatorial online learning based on optimizing feedbacks

Abstract

Combinatorial online learning studies how to learn the unknown parameters and gradually find the optimal combination of targets during the interactions with the environment.This problem has a wide range of applications including advertisement placement, searching and recommendation.Firstly, the definition of combinatorial online learning and its general framework – the problem of combinatorial multi-armed bandits were introduced, and its traditional algorithms and research progress were summarized.Then, the related works of two specific applications, online influence maximization and online learning to rank, were introduced.Finally, the prospective directions of further researches on combinatorial online learning were discussed.

Keywords

• combinatorial multi-armed bandits;online learning;online influence maximization;online learning to rank