A Transferable Edge Caching Method Based On Reinforcement Learning for Dense Small Cell Network

In recent years, data traffic has grown at a drastic speed. Caching popular contents proactively at the edge of networks has become one of the effective methods to offload the huge burden on backhaul links. The reinforcement learning (RL) based edge caching method is capable to fit into the changeable environments and transfer its parameters to other caching cells. In order to converge to the optimal policy quickly and avert the cold-start problem, we propose a transferable edge caching method based on reinforcement learning. The method relies on Asynchronous Advantage Actor-Critic (A3C) algorithm and is applied to dense small cell networks (DSCNs) under content popularity diversity. Compared with Q-Learning based and other caching methods, simulation results verifies that the proposed approach offers faster convergence and efficiently avoids coldstart problem.