Communication-Efficient Algorithm for Asynchronous Multi-Agent Bandits

Xuchuang Wang, Yu-Zhen Janice Chen, Xutong Liu, Lin Yang, Mohammad Hajiesmaili, Don Towsley, John C.S. Lui

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Multi-Agent Multi-Armed Bandits, Distributed Learning, Efficient Communication

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A distributed bandit algorithm for asynchronous multi-agent scenario with constant (time independent) communications and privacy guarantee.

Abstract: We study the cooperative asynchronous multi-agent multi-armed bandits problem, where the active (arm pulling) decision rounds of each agent are asynchronous. In each round, only a subset of agents is active to pull arms, and this subset is unknown and time-varying. We propose a fully distributed algorithm that relies on novel asynchronous communication protocols. This algorithm attains near-optimal regret with constant (time-independent) communications for adversarial asynchronicity among agents. Furthermore, to protect the privacy of the learning process, we extend our algorithms to achieve local differential privacy with rigorous guarantees. Lastly, we report numerical simulations of our new asynchronous algorithms with other known baselines.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7429