Problem
Research questions and friction points this paper is trying to address.
Distributed Algorithms
Multi-Armed Bandit Problems
Robotics Networks
Decentralized Upper Confidence Bound Algorithms for Homogeneous Multi-Agent Multi-Armed Bandits
🤖 AI Summary
This paper studies the decentralized homogeneous multi-agent multi-armed bandit (MAB) problem: $N$ agents collaboratively learn a shared $M$-arm reward distribution via local communication over a fixed graph, without global coordination. We propose the first decentralized UCB1 and KL-UCB algorithms applicable to both undirected and directed graphs, integrating distributed consensus with neighborhood averaging. Theoretically, on undirected graphs, our algorithms achieve an $O(log T)$ asymptotic regret bound; individual regret decreases monotonically with the number of neighbors, yielding a synergistic “whole-greater-than-sum-of-parts” gain. The directed-graph variant also significantly outperforms the single-agent baseline. Extensive experiments validate the effectiveness and robustness of the proposed cooperative learning framework.
📝 Abstract
This paper studies a decentralized homogeneous multi-armed bandit problem in a multi-agent network. The problem is simultaneously solved by $N$ agents assuming they face a common set of $M$ arms and share the same arms' reward distributions. Each agent can receive information only from its neighbors, where the neighbor relationships among the agents are described by a fixed graph. Two fully decentralized upper confidence bound (UCB) algorithms are proposed for undirected graphs, respectively based on the classic algorithm and the state-of-the-art Kullback-Leibler upper confidence bound (KL-UCB) algorithm. The proposed decentralized UCB1 and KL-UCB algorithms permit each agent in the network to achieve a better logarithmic asymptotic regret than their single-agent counterparts, provided that the agent has at least one neighbor, and the more neighbors an agent has, the better regret it will have, meaning that the sum is more than its component parts. The same algorithm design framework is also extended to directed graphs through the design of a variant of the decentralized UCB1 algorithm, which outperforms the single-agent UCB1 algorithm.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Decentralized Decision-making
Multi-Robot Systems
Networked Multi-Armed Bandits
🔎 Similar Papers
No similar papers found.
