Decentralized Task Offloading for Multi-User Mobile Edge Computing with Machine Learning

Abstract

The rapid growth of real-time and computation-intensive applications, such as face recognition, virtual reality, 3D gaming, augmented reality, and intelligent transportation systems, has significantly increased the demand for efficient data processing and low-latency services. However, mobile devices are constrained by limited computational capabilities and battery capacity, making them unsuitable for executing heavy workloads locally. Mobile Edge Computing (MEC) has emerged as a promising paradigm to offload computation tasks to nearby edge servers, thereby reducing latency and improving service quality. Despite its advantages, task offloading in MEC environments remains challenging due to the distributed nature of edge resources, energy constraints of end devices, and dynamic net- work conditions. Existing solutions based on heuristic methods, genetic algorithms, NOMA-based techniques, and mobility-aware services often suffer from high latency, excessive energy consumption, and task migration overhead. To address these limitations, this paper investigates a reinforcement learning-based computation offloading strategy using an improved Deep Deterministic Policy Gradient (IDDPG) algorithm. The proposed IDDPG approach enables decentralized decision-making by learning optimal offloading policies from local observations, effectively balancing local execution and task offloading. By minimizing computation costs, power consumption, and latency, the proposed method outperforms greedy offloading strategies and demonstrates improved efficiency in dynamic MEC environments.