mlsanityData to 14 Oct 2026

Robust Contrastive Learning Using Negative Samples with Diminished Semantics

arXiv 2110.14189 ↗First posted 27 Oct 2021cs.CV cs.LGcited by 43

Not on the map. Why?

Abstract (arXiv)

Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such features. Contrastive learning utilizes positive pairs that preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples in a reversed way, where only the superfluous instead of the semantic features are preserved. We develop two methods, texture-based and patch-based augmentations, to generate negative samples. These samples achieve better generalization, especially under out-of-domain settings. We also analyze our method and the generated texture-based samples, showing that texture features are indispensable in classifying particular ImageNet classes and especially finer classes. We also show that model bias favors texture and shape features differently under different test settings. Our code, trained models, and ImageNet-Texture dataset can be found at https://github.com/SongweiGe/Contrastive-Learning-with-Non-Semantic-Negatives.

Key facts. Cited by 43 papers in the map (data through 14 Oct 2026). Builds on 45 of its 64 references that are in the map. First posted on arXiv 27 Oct 2021.

Builds on

45 of 64 references.

References · 45

  1. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  2. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  3. 2020Bootstrap your own latent: A new approach to self-supervised Learningcited by 4,750
  4. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  5. 2020Exploring Simple Siamese Representation Learningcited by 2,866
  6. 2020Unsupervised Learning of Visual Features by Contrasting Cluster Assignmentscited by 2,857
  7. 2020Improved Baselines with Momentum Contrastive Learningcited by 2,287
  8. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,850
  9. 2018Unsupervised Feature Learning via Non-Parametric Instance-level Discriminationcited by 1,845
  10. 2021Barlow Twins: Self-Supervised Learning via Redundancy Reductioncited by 1,838
  11. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  12. 2016Unsupervised Learning of Visual Representations by Solving Jigsaw Puzzlescited by 1,382
33 more
  1. 2019Contrastive Multiview Codingcited by 1,301
  2. 2019Learning Robust Global Representations by Penalizing Local Predictive Powercited by 1,013
  3. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  4. 2020What Makes for Good Views for Contrastive Learning?cited by 846
  5. 2019Self-Supervised Learning of Pretext-Invariant Representationscited by 808
  6. 2019Data-Efficient Image Recognition with Contrastive Predictive Codingcited by 673
  7. 2020Contrastive Learning with Hard Negative Samplescited by 628
  8. 2017Adversarial Examples for Evaluating Reading Comprehension Systemscited by 571
  9. 2021CutPaste: Self-Supervised Learning for Anomaly Detection and Localizationcited by 531
  10. 2017Adversarial Patchcited by 517
  11. 2020Measuring Robustness to Natural Distribution Shifts in Image Classificationcited by 454
  12. 2020Hard Negative Mixing for Contrastive Learningcited by 402
  13. 2020Debiased Contrastive Learningcited by 356
  14. 2020Are we done with ImageNet?cited by 333
  15. 2019High Frequency Component Helps Explain the Generalization of Convolutional Neural Networkscited by 310
  16. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  17. 2018Generalisation in humans and deep neural networkscited by 237
  18. 2018Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systemscited by 234
  19. 2020What Should Not Be Contrastive in Contrastive Learningcited by 226
  20. 2021Unlearnable Examples: Making Personal Data Unexploitablecited by 193
  21. 2020Adversarial Self-Supervised Contrastive Learningcited by 139
  22. 2020What shapes feature representations? Exploring datasets, architectures, and trainingcited by 126
  23. 2021Can contrastive learning avoid shortcut solutions?cited by 114
  24. 2017Smart Mining for Deep Metric Learningcited by 96
  25. 2020Contrastive Learning with Adversarial Examplescited by 91
  26. 2017Measuring the tendency of CNNs to Learn Surface Statistical Regularitiescited by 87
  27. 2020Hard negative examples are hard, but usefulcited by 82
  28. 2020Shape-Texture Debiased Neural Network Trainingcited by 76
  29. 2021Shape or Texture: Understanding Discriminative Features in CNNscited by 52
  30. 2020Are all negatives created equal in contrastive instance discrimination?cited by 50
  31. 2021Does enhanced shape bias improve neural network robustness to common corruptions?cited by 32
  32. 2018Unsupervised Hard Example Mining from Videos for Improved Object Detectioncited by 30
  33. 2020Learning Visual Representations for Transfer Learning by Suppressing Texturecited by 11

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 16; 2023: 11; 2024: 9; 2025: 5; 2026: 2.

43 papers.

Sort:

By area:

  1. 2022Contrastive Adapters for Foundation Model Group Robustnesscited by 62
  2. 2022Hyperbolic Contrastive Learning for Visual Representations beyond Objectscited by 61
  3. 2022Self-Supervised Learning with an Information Maximization Criterioncited by 38
  4. 2022Contrastive Learning for Unsupervised Domain Adaptation of Time Seriescited by 36
  5. 2022A simple, efficient and scalable contrastive masked autoencoder for learning visual representationscited by 25
  6. 2023HaLP: Hallucinating Latent Positives for Skeleton-based Self-Supervised Learning of Actionscited by 25
  7. 2023Not All Semantics are Created Equal: Contrastive Self-supervised Learning with Automatic Temperature Individualizationcited by 25
  8. 2023PointCMP: Contrastive Mask Prediction for Self-supervised Learning on Point Cloud Videoscited by 22
  9. 2022COCOA: Cross Modality Contrastive Learning for Sensor Datacited by 21
  10. 2023Finding Order in Chaos: A Novel Data Augmentation Method for Time Series in Contrastive Learningcited by 18
  11. 2022Is Self-Supervised Learning More Robust Than Supervised Learning?cited by 17
  12. 2025A Multi-Task Foundation Model for Wireless Channel Representation Using Contrastive and Masked Autoencoder Learningcited by 17
31 more
  1. 2022Training Debiased Subnetworks with Contrastive Weight Pruningcited by 16
  2. 2024Does Negative Sampling Matter? A Review with Insights into its Theory and Applicationscited by 15
  3. 2022Hard Negative Sampling Strategies for Contrastive Representation Learningcited by 13
  4. 2026TaF-VLA: Tactile-Force Alignment in Vision-Language-Action Models for Force-aware Manipulationcited by 13
  5. 2024Reducing Texture Bias of Deep Neural Networks via Edge Enhancing Diffusioncited by 11
  6. 2022Improving Fine-tuning of Self-supervised Models with Contrastive Initializationcited by 10
  7. 2023Sample and Predict Your Latent: Modality-free Sequential Disentanglement via Contrastive Estimationcited by 10
  8. 2024Understanding Hyperbolic Metric Learning through Hard Negative Samplingcited by 10
  9. 2023Affinity Uncertainty-based Hard Negative Mining in Graph Contrastive Learningcited by 9
  10. 2024Learning the Unlearned: Mitigating Feature Suppression in Contrastive Learningcited by 9
  11. 2022Self-supervised debiasing using low rank regularizationcited by 8
  12. 2024Learning Visual Hierarchies in Hyperbolic Space for Image Retrievalcited by 8
  13. 2025SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoningcited by 7
  14. 2024DeNetDM: Debiasing by Network Depth Modulationcited by 6
  15. 2022Certifying Model Accuracy under Distribution Shiftscited by 5
  16. 2022Deeper Insights into the Robustness of ViTs towards Common Corruptionscited by 5
  17. 2023Learning to Count without Annotationscited by 5
  18. 2024SynCo: Synthetic Hard Negatives for Contrastive Visual Representation Learningcited by 5
  19. 2023Towards the Sparseness of Projection Head in Self-Supervised Learningcited by 4
  20. 2024Label-aware Hard Negative Sampling Strategies with Momentum Contrastive Learning for Implicit Hate Speech Detectioncited by 4
  21. 2022Rethinking Rotation in Self-Supervised Contrastive Learning: Adaptive Positive or Negative Data Augmentationcited by 3
  22. 2023Enhancing Contrastive Learning with Noise-Guided Attack: Towards Continual Relation Extraction in the Wildcited by 3
  23. 2025Shape Bias and Robustness Evaluation via Cue Decomposition for Image Classification and Segmentationcited by 3
  24. 2022Robustness of Unsupervised Representation Learning without Labelscited by 2
  25. 2023Pretrained deep models outperform GBDTs in Learning-To-Rank under label scarcitycited by 1
  26. 2023Masking Improves Contrastive Self-Supervised Learning for ConvNets, and Saliency Tells You Wherecited by 1
  27. 2025Discovering Global False Negatives On the Fly for Self-supervised Contrastive Learningcited by 1
  28. 2022Contrastive View Design Strategies to Enhance Robustness to Domain Shifts in Downstream Object Detectioncited by 0
  29. 2024ManiNeg: Manifestation-guided Multimodal Pretraining for Mammography Classificationcited by 0
  30. 2025When Unsupervised Domain Adaptation meets One-class Anomaly Detection: Addressing the Two-fold Unsupervised Curse by Leveraging Anomaly Scarcitycited by 0
  31. 2026Xray-Visual Models: Scaling Vision models on Industry Scale Datacited by 0