ReVFed: Representation-Based Privacy-Preserving Vertical Federated Learning with Heterogeneous Models

Shuo Wang, Jing Yu, Keke Gai*, Liehuang Zhu

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Vertical Federated Learning (VFL) allows institutions to collaborate on machine learning while keeping their original data private. VFL methods currently concentrate on collaborative training among participants with the same model architectures and use cryptography to protect training parameters. In real application scenarios, participants with different computing resources are more likely to select heterogeneous local models to participate in VFL training automatically. However, current methods face challenges in achieving privacy-preserving VFL with heterogeneous participants. To address the above issues, we propose a novel method called Representation-based Privacy-preserving Vertical Federated Learning with Heterogeneous Models (ReVFed). To reduce the impact of the local model on training, we proposed representation aggregation to incorporate each participant’s local knowledge. Furthermore, we also propose a differential privacy-based protection method to protect local feature representations. Experimental results show that ReVFed effectively ensures privacy-preserving training in VFL with heterogeneous models and delivers excellent performance.

Original languageEnglish
Title of host publicationKnowledge Science, Engineering and Management - 17th International Conference, KSEM 2024, Proceedings
EditorsCungeng Cao, Huajun Chen, Liang Zhao, Junaid Arshad, Yonghao Wang, Taufiq Asyhari
PublisherSpringer Science and Business Media Deutschland GmbH
Pages386-397
Number of pages12
ISBN (Print)9789819754977
DOIs
Publication statusPublished - 2024
Event17th International Conference on Knowledge Science, Engineering and Management, KSEM 2024 - Birmingham, United Kingdom
Duration: 16 Aug 202418 Aug 2024

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume14886 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference17th International Conference on Knowledge Science, Engineering and Management, KSEM 2024
Country/TerritoryUnited Kingdom
CityBirmingham
Period16/08/2418/08/24

Keywords

  • Differential privacy
  • Heterogeneous Models
  • Representation learning
  • Vertical federated learning

Fingerprint

Dive into the research topics of 'ReVFed: Representation-Based Privacy-Preserving Vertical Federated Learning with Heterogeneous Models'. Together they form a unique fingerprint.

Cite this