Spectrum and Computing Resource Management for Federated Learning in Distributed Industrial IoT

Federated learning (FL) is a distributed paradigm to support deep neural network (DNN) training while preserving the data owners' privacy. In this paper, we investigate the resource management problem for FL in distributed industrial Internet of Things (IIoT) networks. Specifically, we introduce a three-layer collaborative architecture to support FL. DNNs are trained locally at the selected IIoT devices, and then the DNN model parameters are aggregated by edge servers every FL epoch or by a cloud server every a few FL epochs to update the global DNN model. To enable efficient FL in the resource-limited IIoT networks, judicious computing and spectrum resource allocation is required for training and transmitting the DNN model parameters. Thus, we formulate a joint device selection and resource allocation problem to minimize the FL evaluating loss while satisfying the strict FL epoch delay and devices' energy consumption requirements. Since the decisions of device selection and resource allocation are coupled, we transform the joint optimization problem into a Markov decision process and propose a dynamic resource management scheme based on deep reinforcement learning approaches to efficiently facilitate the FL. Simulation results demonstrate that the proposed scheme can effectively improve the FL performance comparing with benchmarks.