IPFed: Identity protected federated learning for user authentication

TL;DR

IPFed addresses the privacy-accuracy trade-off in federated learning for user authentication by introducing random projection of class embeddings and a separate parameter server to enable secure aggregation. The authors prove that, under an orthonormal transformation $r^t$, IPFed achieves learning equivalence to FedFace and preserves training data privacy. Experiments on CASIA-WebFace with standard benchmarks (LFW, IJB-A, IJB-C) show IPFed matches FedFace's accuracy while keeping class embeddings private, outperforming fixed embedding baselines. The work suggests a practical pathway for privacy-preserving biometric authentication with comparable performance to centralized or server-optimized methods.

Abstract

With the development of laws and regulations related to privacy preservation, it has become difficult to collect personal data to perform machine learning. In this context, federated learning, which is distributed learning without sharing personal data, has been proposed. In this paper, we focus on federated learning for user authentication. We show that it is difficult to achieve both privacy preservation and high accuracy with existing methods. To address these challenges, we propose IPFed which is privacy-preserving federated learning using random projection for class embedding. Furthermore, we prove that IPFed is capable of learning equivalent to the state-of-the-art method. Experiments on face image datasets show that IPFed can protect the privacy of personal data while maintaining the accuracy of the state-of-the-art method.

Figures (2)

• Figure 1: Federated learning for user authentication.
• Figure 2: The overview of IPFed.