Learning and forgetting for server selection in Mobile Edge Computing: a perspective

Abstract

Mobile Edge Computing (MEC) is an architecture that brings computational capabilities to the edge of mobile networks. It enables low-latency and efficient task execution by allowing the Mobile Units (MU) to offload computation tasks to nearby computing servers. However, server selection in MEC remains an open and complex challenge due to the non-stationary nature of the system dynamics, where channel conditions, server loads, and resource availability change over time. Traditional reinforcement learning approaches, while effective for adaptive decision-making, often assume stationary system dynamics. In this perspective, we discuss the role of learning and forgetting mechanisms in MEC server selection, emphasizing the need for adaptive mechanisms that can retain and exploit relevant experience while discarding outdated information.