Publication

Abstract : Supporting computation-intensive and latency-sensitive applications in edge and cloudlet contexts requires effective task offloading. In this research, we propose a cooperative task offloading method based on Soft Actor–Critic (SAC) multi-agent deep reinforcement learning.Under dynamic system settings, several agents work together to learn the best offloading choices while maximizing latency, energy use, and resource use. In continuous action spaces, consistent learning and efficient exploration are made possible by the entropy-based SAC framework. The suggested multi-agent SAC-based framework performs noticeably better than conventional centralized, heuristic, and single-agent reinforcement learning techniques, according to extensive simulations.The outcomes show that cooperative multi-agent SAC is a successful method for scalable task offloading in edge and cloudlet computing systems, resulting in decreased response time and energy consumption along with enhanced load balancing and system performance.