Deep Multi‐Agent Reinforcement Learning for Cooperative Edge Caching
M. Cenk Gursoy, Chen Zhong, Senem Velipasalar · 2019
Wireless edge caching is considered one of the key techniques to reduce data traffic congestion in backhaul links. This chapter is devoted to the application of deep reinforcement learning (DRL) strategies to edge caching at both small base stations and user equipment (UE). It addresses system modelling and problem formulation for different cellular deployment scenarios. The chapter then presents a multi-agent actor-critic DRL framework for edge caching with the goal to increase the cache hit rate and reduce transmission delay. By providing extensive simulation results, it focuses on demonstrating the performance improvements with the proposed DRL policies and provides comparisons with the least recently used, least frequently used, and first-in-first-out caching strategies when caching is performed at small base stations. In terms of working principles and caching performance, the chapter further compares DRL with naive and probabilistic caching policies when caching is performed at UEs.