Federated learning-based computation offloading optimization in edge computing-supported internet of things
Recent visualizations of smart cities, factories, healthcare system and etc. raise challenges on the capability and connectivity of massive Internet of Things (IoT) devices. Hence, edge computing is emerged to complement these capability-constrained devices with an idea offloading intensive computation tasks from them to edge nodes. By taking advantage of this feature, IoT devices are able to conserve more energy and still maintain the quality of services they shall provide. Nevertheless, computation offloading decisions concern joint and complex resource management and should be determined in real time facing dynamic workloads and radio environment. Therefore, in this work, we use multiple Deep Reinforcement Learning (DRL) agents deployed on IoT devices to instruct the decision making of themselves. On the other hand, Federated Learning is utilized to train DRL agents in a distributed fashion, aiming to make the DRL-based decision making practical and further decrease the transmission cost between IoT devices and Edge Nodes. Experimental results corroborate the effectiveness of both the DRL and Federated Learning in the dynamic IoT system.