Clustered Federated Multitask Learning on Non-IID Data With Enhanced Privacy

Federated learning is a machine learning prgadigm that enables the collaborative learning among clients while keeping the privacy of clients’ data. Federated multitask learning (FMTL) deals with the statistic challenge of non-independent and identically distributed (IID) data by training a personali...

Full description

Saved in:
Bibliographic Details
Published in:IEEE internet of things journal Vol. 10; no. 4; pp. 3453 - 3467
Main Authors: Shu, Jiangang, Yang, Tingting, Liao, Xinying, Chen, Farong, Xiao, Yao, Yang, Kan, Jia, Xiaohua
Format: Journal Article
Language:English
Published: Piscataway The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 15.02.2023
Subjects:
Clients
Clustering
Computer architecture
Deep learning
Machine learning
Model accuracy
Numerical methods
Privacy
Security
Training
ISSN:2327-4662, 2327-4662
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Federated learning is a machine learning prgadigm that enables the collaborative learning among clients while keeping the privacy of clients’ data. Federated multitask learning (FMTL) deals with the statistic challenge of non-independent and identically distributed (IID) data by training a personalized model for each client, and yet requires all the clients to be always online in each training round. To eliminate the limitation of full-participation, we explore multitask learning associated with model clustering, and first propose a clustered FMTL to achieve the multual-task learning on non-IID data, while simultaneously improving the communication efficiency and the model accuracy. To enhance its privacy, we adopt a general dual-server architecture and further propose a secure clustered FMTL by designing a series of secure two-party computation protocols. The convergence analysis and security analysis is conducted to prove the correctness and security of our methods. Numeric evaluation on public data sets validates that our methods are superior to state-of-the-art methods in dealing with non-IID data while protecting the privacy.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2327-4662
2327-4662
DOI:10.1109/JIOT.2022.3228893