Title: Non independent data processing for robust federated learning in massive IoT

Abstract: 6G aims to enhance even further 5G capabilities, by accommodating use cases of massive IoT (MIoT) and real-time Machine Learning (ML). However, ML and other branches of it, such as Federated Machine Learning (FML), are highly data sensitive where. Data inaccuracy can cause FML performance degradation and suffer from attacks. Solutions should process the raw data to make it more clean, robust and representative as the FML needs to learn the underlying patterns from data, especially the non-iid data. This work presents a novel approach to address these challenges by giving a ML-based feature selection approach for horizontal federated learning (HFL). First, a personalized novel support vector machine algorithm is designed to detect dirty data and intrusion. Then, the clean data is gathered for the global representative selections by the proposed maximum relevance algorithm, the data will have more global typical values than raw data from MIoT. Finally, based on both two steps, the local IoT gateway could optimize the local data to generate more data for global HFL. The proposed methods can improve the robustness and performance of federated learning, especially when more IoT clients join in the HFL process.

Bio: Xunzheng Zhang received the B.Sc. and M.Sc. degree of Communication Engineering from Shandong University, China. He is currently pursuing the Ph.D. degree at the University of Bristol, UK. He has experience with Internet of Things and edge-cloud collaboration testbed deployment under Industry 4.0. He has the qualification of electronic technology application engineer, owns 4 patents and won the Best Paper Award of IEEE PIMRC 2020, London. His current research interests include 6G massive Internet of Things, federated learning, edge computing and machine learning.