Attention-Driven Deep Reinforcement Learning for Efficient Task Offloading in Mobile Edge Computing

Abstract

Translate Article

Advances in cloud computing and IoT have driven the optimization of distributed systems, enabling near-instantaneous responses. Mobile Edge Computing (MEC) enhances 6 G networks by migrating resource-intensive tasks to edge servers, reducing latency and energy use. This study integrates task offloading and resource allocation to minimize system-wide costs. Although Deep Reinforcement Learning (DRL) is adept at balancing immediate and long-term objectives, resource constraints across mobile devices and servers complicate its application. We propose multi-agent deep reinforcement learning with attention mechanism (AM_MADRL), an attention-based DRL framework, to address these challenges. The algorithm leverages attention mechanisms to extract task features and predict resource demands, enabling MEC servers to make coordinated offloading decisions while respecting system constraints. Simulations confirm AM_MADRL’s superiority over existing methods in cost efficiency and scalability, positioning it as a robust solution for dynamic edge networks.