Alpha Divergence Losses for Biometric Verification

TL;DR

This paper introduces two novel margin-based alpha-divergence loss functions for biometric verification, improving performance and sparsity, and demonstrating significant gains on face and speaker verification benchmarks.

Contribution

The paper proposes two new alpha-divergence loss variants, Q-Margin and A3M, integrating angular margins and addressing training instability, advancing biometric verification methods.

Findings

Significant performance improvements on IJB-B and IJB-C benchmarks.

Strong results in speaker verification on VoxCeleb.

Outperforms baselines at low false acceptance rates.

Abstract

Performance in face and speaker verification is largely driven by margin-based softmax losses such as CosFace and ArcFace. Recently introduced $\alpha$-divergence loss functions offer a compelling alternative, particularly due to their ability to induce sparse solutions (when $\alpha>1$). However, integrating an angular margin-crucial for verification tasks-is not straightforward. We find that this integration can be achieved in at least two distinct ways: via the reference measure (prior probabilities) or via the logits (unnormalized log-likelihoods). In this paper, we explore both pathways, deriving two novel margin-based $\alpha$-divergence losses: Q-Margin (margin in the reference measure) and A3M (margin in the logits). We identify and address a training instability in A3M-caused by sparsity-with a simple yet effective prototype re-initialization strategy. Our methods achieve significant performance gains on the challenging IJB-B and IJB-C face verification benchmarks. We demonstrate similarly strong performance in speaker verification on VoxCeleb. Crucially, our models significantly outperform strong baselines at low false acceptance rates (FAR). This capability is critical for practical high-security applications, such as banking authentication, when minimizing false authentications is paramount. Finally, the sparsity of $\alpha$-divergence-based posteriors enables memory-efficient training, which is crucial for datasets with millions of identities.