Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Meta-Learning for Federated Face Recognition in Imbalanced Data Regimes

Arwin Gansekoele, Emiel Hess, Sandjai Bhulai

TL;DR

This paper addresses privacy in face recognition by proposing Federated Face Recognition (FFR) and extending it with Hessian-Free Model-Agnostic Meta-Learning (HF-MAML) and an embedding regularization term to handle data heterogeneity. It introduces three CelebA-based partitions to simulate non-IID conditions and demonstrates that HF-MAML improves verification performance and fairness, especially when partitions induce quantity and attribute skew. A key contribution is the embedding regularization loss that aligns global and local embeddings, aiding transfer to new clients without relying on a global dataset. The work provides a stepping stone toward privacy-preserving, fair, and personalized face recognition in federated settings and suggests pathways for applying HF-MAML to additional datasets and heterogeneity types.

Abstract

The growing privacy concerns surrounding face image data demand new techniques that can guarantee user privacy. One such face recognition technique that claims to achieve better user privacy is Federated Face Recognition (FRR), a subfield of Federated Learning (FL). However, FFR faces challenges due to the heterogeneity of the data, given the large number of classes that need to be handled. To overcome this problem, solutions are sought in the field of personalized FL. This work introduces three new data partitions based on the CelebA dataset, each with a different form of data heterogeneity. It also proposes Hessian-Free Model Agnostic Meta-Learning (HF-MAML) in an FFR setting. We show that HF-MAML scores higher in verification tests than current FFR models on three different CelebA data partitions. In particular, the verification scores improve the most in heterogeneous data partitions. To balance personalization with the development of an effective global model, an embedding regularization term is introduced for the loss function. This term can be combined with HF-MAML and is shown to increase global model verification performance. Lastly, this work performs a fairness analysis, showing that HF-MAML and its embedding regularization extension can improve fairness by reducing the standard deviation over the client evaluation scores.

Meta-Learning for Federated Face Recognition in Imbalanced Data Regimes

TL;DR

This paper addresses privacy in face recognition by proposing Federated Face Recognition (FFR) and extending it with Hessian-Free Model-Agnostic Meta-Learning (HF-MAML) and an embedding regularization term to handle data heterogeneity. It introduces three CelebA-based partitions to simulate non-IID conditions and demonstrates that HF-MAML improves verification performance and fairness, especially when partitions induce quantity and attribute skew. A key contribution is the embedding regularization loss that aligns global and local embeddings, aiding transfer to new clients without relying on a global dataset. The work provides a stepping stone toward privacy-preserving, fair, and personalized face recognition in federated settings and suggests pathways for applying HF-MAML to additional datasets and heterogeneity types.

Abstract

The growing privacy concerns surrounding face image data demand new techniques that can guarantee user privacy. One such face recognition technique that claims to achieve better user privacy is Federated Face Recognition (FRR), a subfield of Federated Learning (FL). However, FFR faces challenges due to the heterogeneity of the data, given the large number of classes that need to be handled. To overcome this problem, solutions are sought in the field of personalized FL. This work introduces three new data partitions based on the CelebA dataset, each with a different form of data heterogeneity. It also proposes Hessian-Free Model Agnostic Meta-Learning (HF-MAML) in an FFR setting. We show that HF-MAML scores higher in verification tests than current FFR models on three different CelebA data partitions. In particular, the verification scores improve the most in heterogeneous data partitions. To balance personalization with the development of an effective global model, an embedding regularization term is introduced for the loss function. This term can be combined with HF-MAML and is shown to increase global model verification performance. Lastly, this work performs a fairness analysis, showing that HF-MAML and its embedding regularization extension can improve fairness by reducing the standard deviation over the client evaluation scores.
Paper Structure (27 sections, 5 equations, 3 figures, 10 tables, 1 algorithm)

This paper contains 27 sections, 5 equations, 3 figures, 10 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated Face Recognition
  4. Meta Learning
  5. Non-IID Data
  6. Fairness
  7. Methods
  8. Federated HF-MAML for FFR
  9. Global Classification
  10. Local Classification
  11. Embedding regularization
  12. Combined algorithm
  13. Data Partitions
  14. Lognormal class partition
  15. Attribute-based partition
  16. ...and 12 more sections

Figures (3)

  • Figure 1: HF-MAML with equal class partition. TAR@FAR 0.1 results after each communication round with the clients 1-15; before and after tuning with 5 batches.
  • Figure 2: HF-MAML model with equal class partition. TAR@FAR 0.1 results after each communication round with the clients 16-20; before and after tuning with 5 batches.
  • Figure 3: The mean TAR@FAR0.1 per client for FedAvg and HFMAML using the local approach on the attributes partition. Percentage improvement by HFMAML is presented at the end of the bar. Clients with weak performance on average have a greater improvement with HFMAML.