Robust Multi-Agent Bandits Over Undirected Graphs
Abstract
We consider a multi-agent multi-armed bandit setting in which honest agents collaborate over a network to minimize regret but malicious agents can disrupt learning arbitrarily. Assuming the network is the complete graph, existing algorithms incur regret in this setting, where is the number of arms and is the arm gap. For , this improves over the single-agent baseline regret of . In this work, we show the situation is murkier beyond the case of a complete graph. In particular, we prove that if the state-of-the-art algorithm is used on the undirected line graph, honest agents can suffer (nearly) linear regret until time is doubly exponential in and . In light of this negative result, we propose a new algorithm for which the -th agent has regret on any connected and undirected graph, where is the number of 's neighbors who are malicious. Thus, we generalize existing regret bounds beyond the complete graph (where ), and show the effect of malicious agents is entirely local (in the sense that only the malicious agents directly connected to affect its long-term regret).
Cite
@article{arxiv.2203.00076,
title = {Robust Multi-Agent Bandits Over Undirected Graphs},
author = {Daniel Vial and Sanjay Shakkottai and R. Srikant},
journal= {arXiv preprint arXiv:2203.00076},
year = {2023}
}