Sparsity-Inducing Divergence Losses for Biometric Verification

2026-06-30

AI Summary

Existing geometric-margin-based loss functions, such as ArcFace and CosFace, struggle to integrate into the α-divergence framework, thereby limiting the joint optimization of sparsity and discriminability. This work proposes Q-Margin, which introduces—for the first time—a principled margin mechanism within a probabilistic framework by embedding margin penalties directly into the reference measure (i.e., the prior distribution). This design explicitly steers the posterior distribution toward both high sparsity and strong discriminability. The method enables efficient and exact training, significantly outperforming ArcFace and CosFace on IJB-B/C face verification and VoxCeleb speaker verification benchmarks, particularly excelling at low false acceptance rates. Furthermore, Q-Margin scales effectively to recognition tasks involving millions of identities.

Abstract

Performance in face and speaker verification is largely driven by margin-penalty softmax losses such as CosFace and ArcFace. Recently introduced $α$-divergence loss functions offer a compelling alternative, particularly due to their ability to induce sparse solutions (when $α>1$). However, standard geometric margins are designed for the softmax function and do not naturally extend to this generalized probabilistic framework. In this paper we propose Q-Margin, a novel $α$-divergence loss that introduces a principled probabilistic margin. Unlike conventional methods that apply geometric penalties to the logits (unnormalized log-likelihoods), Q-Margin encodes the margin penalty directly into the reference measure (prior probabilities). This formulation naturally encourages discriminative embeddings while preserving the beneficial sparsity properties of the $α$-divergence. We demonstrate that Q-Margin achieves competitive or superior performance on the challenging IJB-B and IJB-C face verification benchmarks and similarly strong results in speaker verification on VoxCeleb. Crucially, against ArcFace and CosFace baselines trained under an identical recipe, Q-Margin consistently improves at low False Acceptance Rates (FARs), a capability critical for practical high-security applications. Finally, the extreme sparsity of the Q-Margin posteriors enables exact and memory-efficient training, offering a scalable solution for datasets with millions of identities.

Problem

• biometric verification
• sparsity-inducing losses
• α-divergence
• probabilistic margin
• face verification

Innovation

• Q-Margin
• α-divergence
• sparsity-inducing loss
• probabilistic margin
• biometric verification