mlsanityData to 14 Oct 2026

A Non-isotropic Probabilistic Take on Proxy-based Deep Metric Learning

arXiv 2207.03784 ↗First posted 8 Jul 2022cs.LG stat.MLcited by 19

Not on the map. Why?

Abstract (arXiv)

Proxy-based Deep Metric Learning (DML) learns deep representations by embedding images close to their class representatives (proxies), commonly with respect to the angle between them. However, this disregards the embedding norm, which can carry additional beneficial context such as class- or image-intrinsic uncertainty. In addition, proxy-based DML struggles to learn class-internal structures. To address both issues at once, we introduce non-isotropic probabilistic proxy-based DML. We model images as directional von Mises-Fisher (vMF) distributions on the hypersphere that can reflect image-intrinsic uncertainties. Further, we derive non-isotropic von Mises-Fisher (nivMF) distributions for class proxies to better represent complex class-specific variances. To measure the proxy-to-image distance between these models, we develop and investigate multiple distribution-to-point and distribution-to-distribution metrics. Each framework choice is motivated by a set of ablational studies, which showcase beneficial properties of our probabilistic approach to proxy-based DML, such as uncertainty-awareness, better-behaved gradients during training, and overall improved generalization performance. The latter is especially reflected in the competitive performance on the standard DML benchmarks, where our approach compares favorably, suggesting that existing proxy-based DML can significantly benefit from a more probabilistic treatment. Code is available at github.com/ExplainableML/Probabilistic_Deep_Metric_Learning.

Key facts. Cited by 19 papers in the map (data through 14 Oct 2026). Builds on 56 of its 82 references that are in the map. First posted on arXiv 8 Jul 2022.

Builds on

56 of 82 references.

References · 56

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  3. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  4. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  5. 2014Going Deeper with Convolutionscited by 4,241
  6. 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,849
  7. 2020Supervised Contrastive Learningcited by 2,829
  8. 2020Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hyperspherecited by 1,405
  9. 2017SphereFace: Deep Hypersphere Embedding for Face Recognitioncited by 737
  10. 2015Deep Metric Learning via Lifted Structured Feature Embeddingcited by 449
  11. 2020Circle Loss: A Unified Perspective of Pair Similarity Optimizationcited by 403
  12. 2019Multi-Similarity Loss with General Pair Weighting for Deep Metric Learningcited by 368
38 more
  1. 2020A Metric Learning Reality Checkcited by 321
  2. 2017Sampling Matters in Deep Embedding Learningcited by 285
  3. 2017No Fuss Distance Metric Learning using Proxiescited by 226
  4. 2021Contrastive Learning Inverts the Data Generating Processcited by 221
  5. 2020Proxy Anchor Loss for Deep Metric Learningcited by 195
  6. 2018Hyperspherical Variational Auto-Encoderscited by 189
  7. 2019Probabilistic Face Embeddingscited by 173
  8. 2017Beyond triplet loss: a deep quadruplet network for person re-identificationcited by 172
  9. 2019SoftTriple Loss: Deep Metric Learning Without Triplet Samplingcited by 163
  10. 2021Probabilistic Embeddings for Cross-Modal Retrievalcited by 151
  11. 2017Deep Metric Learning with Angular Losscited by 107
  12. 2020ProxyNCA++: Revisiting and Revitalizing Proxy Neighborhood Component Analysiscited by 107
  13. 2017Multi-Level Variational Autoencoder: Learning Disentangled Representations from Grouped Observationscited by 100
  14. 2020Revisiting Training Strategies and Generalization Performance in Deep Metric Learningcited by 100
  15. 2020A unifying mutual information view of metric learning: cross-entropy vs. pairwise lossescited by 100
  16. 2017L2-constrained Softmax Loss for Discriminative Face Verificationcited by 96
  17. 2017Smart Mining for Deep Metric Learningcited by 96
  18. 2020Rethinking Zero-shot Video Classification: End-to-end Training for Realistic Applicationscited by 80
  19. 2019Improved Embeddings with Easy Positive Triplet Miningcited by 67
  20. 2019Hardness-Aware Deep Metric Learningcited by 55
  21. 2017von Mises-Fisher Mixture Model-based Deep learning: Application to Face Verificationcited by 46
  22. 2021von Mises-Fisher Loss: An Exploration of Embedding Geometries for Supervised Learningcited by 42
  23. 2019MIC: Mining Interclass Characteristics for Improved Metric Learningcited by 41
  24. 2018Deep Metric Learning with BIER: Boosting Independent Embeddings Robustlycited by 40
  25. 2020DiVA: Diverse Visual Feature Aggregation for Deep Metric Learningcited by 40
  26. 2019Divide and Conquer the Embedding Space for Metric Learningcited by 39
  27. 2020Fewer is More: A Deep Graph Metric Learning Perspective Using Fewer Proxiescited by 36
  28. 2022Non-isotropy Regularization for Proxy-based Deep Metric Learningcited by 33
  29. 2019Metric Learning With HORDE: High-Order Regularizer for Deep Embeddingscited by 29
  30. 2018Von Mises-Fisher Loss for Training Sequence to Sequence Models with Continuous Outputscited by 27
  31. 2021Deep Relational Metric Learningcited by 22
  32. 2021Embedding Transfer with Label Relaxation for Improved Metric Learningcited by 21
  33. 2021Learning with Memory-based Virtual Classes for Deep Metric Learningcited by 21
  34. 2019The Group Loss for Deep Metric Learningcited by 20
  35. 2020PADS: Policy-Adapted Sampling for Visual Similarity Learningcited by 20
  36. 2022Integrating Language Guidance into Vision-based Deep Metric Learningcited by 19
  37. 2021Characterizing Generalization under Out-Of-Distribution Shifts in Deep Metric Learningcited by 17
  38. 2018Directional Statistics-based Deep Metric Learning for Image Classification and Retrievalcited by 16

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 3; 2024: 9; 2025: 2; 2026: 5.

19 papers.

Sort:

By area:

  1. 2023Waffling around for Performance: Visual Classification with Random Words and Broad Conceptscited by 101
  2. 2024InfoNCE: Identifying the Gap Between Theory and Practicecited by 40
  3. 2024Cross-Entropy Is All You Need To Invert the Data Generating Processcited by 31
  4. 2023Probabilistic Contrastive Learning Recovers the Correct Aleatoric Uncertainty of Ambiguous Inputscited by 29
  5. 2023URL: A Representation Learning Benchmark for Transferable Uncertainty Estimatescited by 17
  6. 2025FreqDebias: Towards Generalizable Deepfake Detection via Consistency-Driven Frequency Debiasingcited by 14
  7. 2024Pretrained Visual Uncertaintiescited by 10
  8. 2025On the Importance of Embedding Norms in Self-Supervised Learningcited by 9
  9. 2024Potential Field Based Deep Metric Learningcited by 4
  10. 2024Context-Aware Multimodal Pretrainingcited by 3
  11. 2024Dynamic Feature Learning and Matching for Class-Incremental Learningcited by 1
  12. 2024Probabilistic Contrastive Learning with Explicit Concentration on the Hyperspherecited by 0
7 more
  1. 2024Realigned Softmax Warping for Deep Metric Learningcited by 0
  2. 2024Reducing Semantic Ambiguity In Domain Adaptive Semantic Segmentation Via Probabilistic Prototypical Pixel Contrastcited by 0
  3. 2026Beyond Seen Bounds: Class-Centric Polarization for Single-Domain Generalized Deep Metric Learningcited by 0
  4. 2026BLENDER: Blended Text Embeddings and Diffusion Residuals for Intra-Class Image Synthesis in Deep Metric Learningcited by 0
  5. 2026Learning to Align Generative Appearance Priors for Fine-grained Image Retrievalcited by 0
  6. 2026Optimization Dynamics Imprint Semantic Specificity in Contrastive Embedding Normscited by 0
  7. 2026Absorbing Gradient Conflicts: Modeling Semantic Variance via Kent Distributions for Cross-Modal Hashingcited by 0