Distributed Learning-based Matching for Task Offloading in Dynamic Fog Computing Networks

초록

Task offloading in dynamic fog computing networks (FCNs) presents significant challenges due to continuously changing task requirements and fluctuating computing resources. This paper proposes Distributed Learning-based Matching (DLMATCH) framework that enables adaptive task offloading through a multi-stage, one-to-many matching mechanism. In DL-MATCH, task nodes (TNs) employ distributed learning to estimate the acceptance probability of helper nodes (HNs) based on historical interactions. By leveraging multi-stage interaction and reward-based learning, DL-MATCH efficiently adapts to unknown preferences on both sides, optimizing task allocation while minimizing offloading delays. Simulation results demonstrate that DL-MATCH outperforms baseline approaches in terms of task completion rate and resource utilization, making it a promising solution for dynamic task offloading in fog computing environments.