Dissertation Title: Efficient Federated Learning Algorithms for Intelligent Edge Devices

Date: 2023/08/29 - 2023/08/29

Dissertation Title: Efficient Federated Learning Algorithms for Intelligent Edge Devices

Speaker: Xiaochen Zhou, Ph.D. candidate at UM-SJTU Joint Institute

Time: August 29th from 10:00 a.m., 2023 (Beijing Time)

Location: Room 403, Longbin Building

Abstract

The prosperity of wireless networks brings massive edge devices to the Internet, facilitating various intelligent applications such as smart manufacturing. In these applications, edge devices need to execute machine learning (ML) models locally for some time-sensitive services such as anomaly detection. Training such ML models requires the data generated by edge devices. However, centralized learning in the cloud breaches users’ data privacy while local learning on edge devices cannot obtain models with high generalization performance. To tackle this dilemma, federated learning (FL) is a promising learning paradigm. In this dissertation, four research topics are identified to resolve some key issues of FL over edge devices: 1) memory-efficient FL for kernel k-means; 2) memory-efficient FL for kernel SVM; 3) domain adaptation for compact models on edge devices; 4) FL over noisy data.