Collaborative Multi-Agent Reinforcement Learning of Caching Optimization in Small-Cell Networks.
IEEE Global Communications Conference(2018)
摘要
Previous works on learning-based caching problems often only focus on a single base station (BS) scenario. For works considering multiple BSs, the coordination among BSs only exists in the cache placement phase after the learning phase that each BS estimates the file popularity independently. In this work, we investigate the cache strategy design problem in small cell networks (SCNs) with multiple small base stations (SBSs) when user preferences are unknown. We model this multi-agent decision making problem in a multi-armed bandit (MAB) perspective. We first tackle this problem with the centralized MAB algorithm and distributed multi-agent MAB (MAMAB) algorithm. For the centralized algorithm, it considers the coordinations among SBSs but the computational complexity grows exponentially with the number of SBSs. For the distributed one, the computational complexity grows linearly with the number of SBSs but the coordinations among SBSs are totally ignored. To take both coordination among SBSs and computational complexity into account, we propose a collaborative MAMAB algorithm to learn the cache strategy directly, rather than learning the user preferences. The simulation results show that the collaborative MAMAB approaches the performance of the greedy algorithm when the user preferences are perfectly known.
更多查看译文
关键词
multiagent reinforcement learning,caching optimization,small-cell networks,learning-based caching problems,single base station scenario,cache placement phase,cache strategy design problem,multiple small base stations,user preferences,multiagent decision making problem,multiarmed bandit perspective,centralized MAB algorithm,multiagent MAB algorithm,computational complexity,collaborative MAMAB algorithm,greedy algorithm,SBS
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络