Combinatorial multi-armed bandits in cognitive radio networks: A brief overview

摘要

Combinatorial multi-armed bandit (MAB) problem can be used to formulate sequential decision problems with exploration-exploitation tradeoff. Dynamic spectrum access (DSA) in cognitive radio (CR) networks is one of important applications. In this work, we briefly overview combinatorial MAB problems with its possible applications to CR networks. We first investigate the standard MAB problems where a single player either explores an arm to gather information to improve its decision strategy, or exploits the arm based on the information that it has collected at each round. Then, we study the taxonomy of combinatorial MAB problems, in particular for multi-player scenarios with independent and identically distributed (i.i.d.) rewards. Finally, we discuss limitations of existing works and interesting open problems.