TY - GEN
T1 - ReVFed
T2 - 17th International Conference on Knowledge Science, Engineering and Management, KSEM 2024
AU - Wang, Shuo
AU - Yu, Jing
AU - Gai, Keke
AU - Zhu, Liehuang
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.
PY - 2024
Y1 - 2024
N2 - 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.
AB - 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.
KW - Differential privacy
KW - Heterogeneous Models
KW - Representation learning
KW - Vertical federated learning
UR - http://www.scopus.com/inward/record.url?scp=85200743129&partnerID=8YFLogxK
U2 - 10.1007/978-981-97-5498-4_30
DO - 10.1007/978-981-97-5498-4_30
M3 - Conference contribution
AN - SCOPUS:85200743129
SN - 9789819754977
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 386
EP - 397
BT - Knowledge Science, Engineering and Management - 17th International Conference, KSEM 2024, Proceedings
A2 - Cao, Cungeng
A2 - Chen, Huajun
A2 - Zhao, Liang
A2 - Arshad, Junaid
A2 - Wang, Yonghao
A2 - Asyhari, Taufiq
PB - Springer Science and Business Media Deutschland GmbH
Y2 - 16 August 2024 through 18 August 2024
ER -