Graduate Seminar: Secure Distributed Machine Learning with Multi Aggregators in Edge Computing

Abstract

Federated learning (FL) is a sought-after distributed machine learning architecture and has been seamlessly integrated into edge computing. However, the limited resources on edge devices introduce security vulnerabilities to FL. While blockchain technology promises to bolster security, edge devices cannot afford the cost of off-the-shelf blockchain systems. Moreover, the Blockchain-FL with multiple aggregators is still under-explored. In this talk, I introduce a novel Blockchain-empowered Heterogeneous Multi-Aggregator Federated Learning Architecture (BMA-FL). We design a performance-based Byzantine consensus mechanism to enable secure and fast model aggregation. I will also discuss the model damaging problem in BMA-FL due to the heterogeneity of aggregators. We propose a multi-agent deep reinforcement learning algorithm to help aggregators decide the best training strategies to mitigate the damage. The experiments on real-word datasets demonstrate the efficiency of BMA-FL to achieve better model performance faster than baselines.

Speaker Biography

Xiao Li is a Ph.D. Candidate in Computer Science at The University of Texas at Dallas advised by Prof. Weili Wu. His research focuses on blockchain technology and its applications in distributed systems. He is also interested in applied data science for problem solving with application of machine learning, deep learning and reinforcement Learning. Xiao has published 10 papers in related fields at refereed conferences and journals such as IEEE ICDCS, IEEE TCSS, Theoretical Computer Science, and Journal of Combinatorial Optimization. Xiao was awarded the prestigious Jan Van der Ziel Fellowship at UT Dallas in 2023. Xiao is also a member of IEEE, ACM and AAAI. Xiao has served as peer reviewers in various reputable conferences and journals including KDD, ICDCS, IJCAI and Information Sciences.