.png){kind=icon}
.png){kind=icon}
Return
EPRA-VFL: A privacy-preserving and efficient verifiable federated learning scheme with robust aggregation
.png){kind=icon}
PRE
AI
L
S
G
G
Z
Y
DOI：10.1016/j.jpdc.2026.105266.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As data privacy and security regulations become increasingly stringent, traditional centralized machine learning methods have faced significant challenges. Federated learning is an emerging distributed learning paradigm that preserves data within local environments and enables multiple clients to collaboratively train shared models, while effectively mitigating privacy risks. However, the federated learning framework is not foolproof and encounters several security issues, including malicious client model pollution attacks, malicious inference attacks on transmitted models, and dishonest servers failing to aggregate models according to the regulations. These threats may involve sending erroneous or harmful model updates to disrupt global model performance, inducing the model to learn incorrect patterns, compromising local client privacy, or issuing incorrect aggregated models. To address these challenges, this study proposes a robust and verifiable federated learning scheme, EPRA-VFL, which integrates local differential privacy, linear homomorphic hashing, and secret sharing technologies. EPRA-VFL enhances defense mechanisms against malicious client attacks while ensuring data privacy and verifying the security and authenticity of server-issued results. Specifically, the single-server architecture combined with linear homomorphic hashing and secret sharing provides a novel solution to ensure both robustness and verifiability in a federated learning environment. The security, trustworthiness, and privacy of the protocol were analyzed, and the efficiency of the scheme and model performance were compared. The results demonstrate that the proposed scheme exhibits high efficiency and robustness against malicious client pollution attacks. This study highlights the synergy between differential privacy, robust aggregation, and verifiability, showcasing the potential of EPRA-VFL as a comprehensive solution for secure and private federated learning.
Keywords:
Federated learning
Privacy protection
Robust aggregation
Verifiability
Journal
IF:
4
Papers:
3.8K
Citations:
4.8K
Organization
Cited Papers