Improving Federated Learning Face Recognition
via Privacy-Agnostic Clusters

ICLR 2022, Spotlight

Qiang Meng1   Feng Zhou1   Hainan Ren1   Tianshu Feng2   Guochao Liu1   Yuanqing Lin1

1Algorithm Research, Aibee Inc.      2Independent Researcher

Overview of the proposed method. Under the federated setting, multiple clients communicate non-sensitive model parameters Φc (excluding the last fully connected layer Wc which are greatly tied to privacy) under the orchestration by a central server. (a) Since Wc's are kept locally in conventional FL updating, the embedding space could overlap for different classes during training. (b) In contrast, the proposed PrivacyFace framework learns an improved face embedding by aggregating discriminative embedding clusters that are proved to achieve differential privacy.


The growing public concerns on data privacy in face recognition can be greatly addressed by the federated learning (FL) paradigm. However, conventional FL methods perform poorly due to the uniqueness of the task: broadcasting class centers among clients is crucial for recognition performances but leads to privacy leakage. To resolve the privacy-utility paradox, this work proposes PrivacyFace, a framework largely improves the federated learning face recognition via communicating auxiliary and privacy-agnostic information among clients. PrivacyFace mainly consists of two components: First, a practical Differentially Private Local Clustering (DPLC) mechanism is proposed to distill sanitized clusters from local class centers. Second, a consensus-aware recognition loss subsequently encourages global consensuses among clients, which ergo results in more discriminative features. The proposed framework is mathematically proved to be differentially private, introducing a lightweight overhead as well as yielding prominent performance boosts (e.g., +9.63% and +10.26% for TAR@FAR=1e-4 on IJB-B and IJB-C respectively). Extensive experiments and ablation studies on a large-scale dataset have demonstrated the efficacy and practicability of our method.


Improving Federated Learning Face Recognition via Privacy-Agnostic Clusters
Qiang Meng1, Feng Zhou1, Hainan Ren1, Tianshu Feng2, Guochao Liu1, Yuanqing Lin1
1Algorithm Research, Aibee Inc.    2Independent Researcher
2022 International Conference on Learning Representations
[paper] [arXiv] [Video-中文] [Poster] [Slides]


Coming soon...

    title={Improving Federated Learning Face Recognition via Privacy-Agnostic Clusters},
    author={Qiang Meng and Feng Zhou and Hainan Ren and Tianshu Feng and Guochao Liu and Yuanqing Lin},
    booktitle={International Conference on Learning Representations},