Enabling Cooperative Relay Selection by Transfer Learning for the Industrial Internet of Things

Large manufacturing sites with movable obstacles and dynamic network topology call for reliable and efficient strategies to transmit data through the industrial Internet of Things (IoT). Cooperative communications and relay selection have shown a great potential to improve throughput and energy efficiency at the expanse of high end-to-end transmission latency. To reduce this latency, we propose to use transfer learning for relay selection in the industrial IoT. Unlike traditional approaches that are trained for a specific task, transfer learning exploits the acquired knowledge from similar tasks to assist new tasks. Transfer learning is capable of improving learning performance, reducing the need for large datasets for different setups, lowering communication overhead and computational complexity. Specifically, in this paper, we propose a generic transfer learning framework for relay selection problems in the industrial IoT. Based on the proposed framework, we design a hypothesis and test it by empirical data for convergence analysis. Also, we devise and conduct a step-by-step and rigorous hyperparameter tuning procedure for the proposed transfer learning framework. The accuracy of the proposed approach is evaluated and verified by extensive training and test datasets abiding by different statistical distributions.