Communication-Efficient Collaborative Best Arm Identification

TL;DR

This paper addresses multi-agent bandit problems focusing on identifying the top arms efficiently through collaboration, emphasizing minimizing communication to maximize learning speedup, supported by theoretical and experimental results.

Contribution

It introduces new algorithms and impossibility results for collaborative top-m arm identification with minimized communication costs.

Findings

Algorithms achieve significant speedup over single-agent methods.

Communication-efficient algorithms outperform existing approaches.

Experimental results validate theoretical advantages.

Abstract

We investigate top-$m$ arm identification, a basic problem in bandit theory, in a multi-agent learning model in which agents collaborate to learn an objective function. We are interested in designing collaborative learning algorithms that achieve maximum speedup (compared to single-agent learning algorithms) using minimum communication cost, as communication is frequently the bottleneck in multi-agent learning. We give both algorithmic and impossibility results, and conduct a set of experiments to demonstrate the effectiveness of our algorithms.