Sparsity-Inducing Divergence Losses for Biometric Verification

• URL: http://arxiv.org/abs/2606.31664v1
• Date: Tue, 30 Jun 2026 13:42:41 GMT
• Title: Sparsity-Inducing Divergence Losses for Biometric Verification
• Abstract summary: We propose Q-Margin, a $$-divergence loss that introduces a principled probabilistic margin.<n>Q-Margin encodes the margin penalty directly into the reference measure (prior probabilities)<n>We demonstrate that Q-Margin achieves competitive or superior performance on the challenging IJB-B and IJB-C face verification benchmarks.
• Score: 15.64389037529846
• License: http://creativecommons.org/licenses/by/4.0/
• 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.

Related papers

• High-Probability Convergence in Decentralized Stochastic Optimization with Gradient Tracking [69.90407799170687]
  We study high-probability (HP) convergence guarantees in decentralized optimization.<n>Our results establish the same conditions on the horizon as in achieving comparable times.
  arXiv  Detail & Related papers  (2026-04-30T22:45:21Z)
• ZIP-RC: Optimizing Test-Time Compute via Zero-Overhead Joint Reward-Cost Prediction [57.799425838564]
  We present ZIP-RC, an adaptive inference method that equips models with zero-overhead inference-time predictions of reward and cost.<n> ZIP-RC improves accuracy by up to 12% over majority voting at equal or lower average cost.
  arXiv  Detail & Related papers  (2025-12-01T09:44:31Z)
• Alpha Divergence Losses for Biometric Verification [19.758259380263528]
  We show that $$-divergence loss functions offer a compelling alternative to margin-based softmax losses.<n>We derivation two novel margin-based $$-divergence losses: Q-Margin and A3M.<n>Our models significantly outperform strong baselines at low false acceptance rates.
  arXiv  Detail & Related papers  (2025-11-17T17:27:28Z)
• Improved High-probability Convergence Guarantees of Decentralized SGD [74.39742894097348]
  We show that $mathttDSGD$ converges in HP under the same conditions on the cost as in the mean-squared error (MSE) sense.<n>Our improved analysis yields linear-up in the number of users, demonstrating that $mathttDSGD$ maintains performance in the HP sense.
  arXiv  Detail & Related papers  (2025-10-07T17:15:08Z)
• $ε$-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise [99.91399796174602]
  Noisy labels pose a common challenge for training accurate deep neural networks.<n>We propose $epsilon$-softmax, which modifies the outputs of the softmax layer to approximate one-hot vectors with a controllable error.<n>We prove theoretically that $epsilon$-softmax can achieve noise-tolerant learning with controllable excess risk bound for almost any loss function.
  arXiv  Detail & Related papers  (2025-08-04T13:10:48Z)
• LH2Face: Loss function for Hard High-quality Face [2.9307424859610447]
  Most face recognition algorithms are based on cosine similarity with softmax classification.<n>A novel loss function is proposed, named Loss function for Hard High-quality Face (LH2Face)<n>Our LH2Face is superior to schemes on hard high-quality face datasets, achieving 49.39% accuracy on the IJB-B dataset.
  arXiv  Detail & Related papers  (2025-06-30T06:59:02Z)
• Expressive Losses for Verified Robustness via Convex Combinations [67.54357965665676]
  We study the relationship between the over-approximation coefficient and performance profiles across different expressive losses. We show that, while expressivity is essential, better approximations of the worst-case loss are not necessarily linked to superior robustness-accuracy trade-offs.
  arXiv  Detail & Related papers  (2023-05-23T12:20:29Z)
• Learning Towards the Largest Margins [83.7763875464011]
  Loss function should promote the largest possible margins for both classes and samples. Not only does this principled framework offer new perspectives to understand and interpret existing margin-based losses, but it can guide the design of new tools.
  arXiv  Detail & Related papers  (2022-06-23T10:03:03Z)
• ElasticFace: Elastic Margin Loss for Deep Face Recognition [6.865656740940772]
  Learning discriminative face features plays a major role in building high-performing face recognition models. Recent state-of-the-art face recognition solutions proposed to incorporate a fixed penalty margin on classification loss function, softmax loss. We propose elastic margin loss (ElasticFace) that allows flexibility in the push for class separability.
  arXiv  Detail & Related papers  (2021-09-20T10:31:50Z)
• Max-Margin is Dead, Long Live Max-Margin! [87.90853526750716]
  We show that Max-Margin loss can only be consistent to the classification task under highly restrictive assumptions on the discrete loss measuring the error between distances. We introduce Restricted-Max-Margin, where the loss-augmented scores is maintained, but performed over a subset of the original domain.
  arXiv  Detail & Related papers  (2021-05-31T15:55:52Z)
• Reliable Probabilistic Face Embeddings in the Wild [5.094757968178221]
  Probabilistic Face Embeddings (PFE) can improve face recognition performance in unconstrained scenarios. PFE methods tend to be over-confident in estimating uncertainty and is too slow to apply to large-scale face matching. This paper proposes a regularized probabilistic face embedding method to improve the robustness and speed of PFE.
  arXiv  Detail & Related papers  (2021-02-08T09:27:57Z)
• Being Bayesian about Categorical Probability [6.875312133832079]
  We consider a random variable of a categorical probability over class labels. In this framework, the prior distribution explicitly models the presumed noise inherent in the observed label. Our method can be implemented as a plug-and-play loss function with negligible computational overhead.
  arXiv  Detail & Related papers  (2020-02-19T02:35:32Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.