publications

2020

1. Edge-Assisted Federated Learning: An Empirical Study from Software Decomposition Perspective

   Shi, Yimin, Duan, Haihan, Chi, Yuanfang, Gai, Keke, and Cai, Wei

   In International Conference on Algorithms and Architectures for Parallel Processing 2020

   Abs HTML PDF

   Federated learning is considered to be a privacy-preserving collaborative machine learning training method. However, due to the general limitation of the computing ability of the terminal device, the training efficiency becomes an issue when training some complex deep neural network models. On the other hand, edges, the nearby stationary devices with higher computational capacity, might serve as a help. This paper presents the design of a component-based federated learning framework, which facilitates the offloading of training layers to nearby edge devices while preserving the users’ privacy. We conduct an empirical study on a classic convolutional neural network to validate our framework. Experiments show that this method can effectively shorten the time cost for mobile terminals to perform local training in the federated learning process.

2022

1. An Energy-efficient and Privacy-aware Decomposition Framework for Edge-assisted Federated Learning

   Shi, Yimin, Duan, Haihan, Yang, Lei, and Cai, Wei

   ACM Transactions on Sensor Networks (TOSN) 2022

   Abs HTML PDF

   Deep Learning (DL) is an essential technology for modern intelligent sensor network and interactive multimedia applications, having problems with user data privacy when training on a central cloud. While Federated Learning (FL) motivates to preserve user privacy, it also causes new problems of lower user terminal usability and training efficiency, which caused substantial energy consumption. This paper proposes a novel energy-efficient and privacy-aware decomposition framework to improve user-side FL efficiency under pre-defined privacy requirements with the assistance of Mobile Edge Computing (MEC) and Software Decomposition. It takes the propagation of each neural layer as the migrating unit and considers the trade-off relationship between privacy and efficiency. We also propose an online scheduling algorithm to optimize the framework’s training performance. Furthermore, we summarize eight privacy-sensitive information classes on which existing privacy attacks base and design configurable privacy preservation mechanisms for each class. Simulations and experiments prove the effectiveness of our framework and algorithm in FL efficiency improvement and the effects of different privacy constraints on the overall training efficiency.