arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2102.08462cs.LGcs.AIcs.MA
收藏

通信受限的多智能体多武装匪徒

Multi-Agent Multi-Armed Bandits with Limited Communication

Purdue University(普渡大学)

Mridul Agarwal, Vaneet Aggarwal, Kamyar Azizzadenesheli

摘要:

我们考虑的问题,$N$代理合作互动的一个随机$K$手臂土匪问题的实例$K \gg N$。代理的目标是同时最小化所有代理的总$T$时间步长,通信轮数,以及在每个通信轮的位数的累积遗憾。我们提出了有限的通信协作-置信上限(LCC-UCB),一个基于加倍历元的算法,每个代理只在历元结束后进行通信,并共享它所知道的最佳手臂的索引。在我们的算法LCC-UCB中,每个代理享有一个$\tilde{O}\left(\sqrt{({K/N}+ N)T}\right)$的遗憾,每个代理进行$O(\log T)$步的通信,并在每个通信步中广播$O(\log K)$比特。我们将这一工作推广到具有最大度$K_G$和直径$D$的稀疏图,并提出了具有$\tilde{O}\left(D\sqrt{(K/N+ K_G)DT}\right)$遗憾界的LCC-UCB-GRAPH.最后,我们的经验表明,LCC-UCB和LCC-UCB-GRAPH算法表现良好,优于通过中心节点通信的策略

英文摘要:

We consider the problem where $N$ agents collaboratively interact with an instance of a stochastic $K$ arm bandit problem for $K \gg N$. The agents aim to simultaneously minimize the cumulative regret over all the agents for a total of $T$ time steps, the number of communication rounds, and the number of bits in each communication round. We present Limited Communication Collaboration - Upper Confidence Bound (LCC-UCB), a doubling-epoch based algorithm where each agent communicates only after the end of the epoch and shares the index of the best arm it knows. With our algorithm, LCC-UCB, each agent enjoys a regret of $\tilde{O}\left(\sqrt{({K/N}+ N)T}\right)$, communicates for $O(\log T)$ steps and broadcasts $O(\log K)$ bits in each communication step. We extend the work to sparse graphs with maximum degree $K_G$, and diameter $D$ and propose LCC-UCB-GRAPH which enjoys a regret bound of $\tilde{O}\left(D\sqrt{(K/N+ K_G)DT}\right)$. Finally, we empirically show that the LCC-UCB and the LCC-UCB-GRAPH algorithm perform well and outperform strategies that communicate through a central node

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑