mlsanityData to 14 Oct 2026

Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations

arXiv 2204.02937 ↗First posted 6 Apr 2022cs.LG cs.CV stat.MLcited by 381

Not on the map. Why?

Abstract (arXiv)

Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significantly reduce reliance on background and texture information, improving robustness to covariate shift, after only minutes of training on a single GPU.

Key facts. Cited by 381 papers in the map (data through 14 Oct 2026). Builds on 100 of its 148 references that are in the map. First posted on arXiv 6 Apr 2022.

Builds on

100 of 148 references.

References · 100

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  4. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  5. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  6. 2017Mask R-CNNcited by 7,277
  7. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  8. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,349
  9. 2021On the Opportunities and Risks of Foundation Modelscited by 4,033
  10. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  11. 2015Domain-Adversarial Training of Neural Networkscited by 3,044
  12. 2015Fast R-CNNcited by 2,933
38 more
  1. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  2. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,218
  3. 2016Equality of Opportunity in Supervised Learningcited by 2,007
  4. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  5. 2014Unsupervised Domain Adaptation by Backpropagationcited by 1,917
  6. 2019Invariant Risk Minimizationcited by 1,854
  7. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,850
  8. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  9. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,375
  10. 2018On the Spectral Bias of Neural Networkscited by 1,341
  11. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,247
  12. 2020In Search of Lost Domain Generalizationcited by 1,054
  13. 2019Learning Imbalanced Datasets with Label-Distribution-Aware Margin Losscited by 989
  14. 2017Revisiting Unreasonable Effectiveness of Data in Deep Learning Eracited by 919
  15. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  16. 2019Decoupling Representation and Classifier for Long-Tailed Recognitioncited by 809
  17. 2019Big Transfer (BiT): General Visual Representation Learningcited by 769
  18. 2020Out-of-Distribution Generalization via Risk Extrapolation (REx)cited by 742
  19. 2022Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distributioncited by 739
  20. 2018Learning to Reweight Examples for Robust Deep Learningcited by 675
  21. 2016Fair prediction with disparate impact: A study of bias in recidivism prediction instrumentscited by 672
  22. 2018Recognition in Terra Incognitacited by 658
  23. 2013Domain Generalization via Invariant Feature Representationcited by 623
  24. 2018Exploring the Limits of Weakly Supervised Pretrainingcited by 605
  25. 2018A Reductions Approach to Fair Classificationcited by 587
  26. 2016Inherent Trade-Offs in the Fair Determination of Risk Scorescited by 586
  27. 2018Do Better ImageNet Models Transfer Better?cited by 565
  28. 2015Causal inference using invariant prediction: identification and confidence intervalscited by 536
  29. 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 523
  30. 2019Learning the Difference that Makes a Difference with Counterfactually-Augmented Datacited by 439
  31. 2019A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmarkcited by 425
  32. 2017On Fairness and Calibrationcited by 406
  33. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 404
  34. 2018Confounding variables can degrade generalization performance of radiological deep learning modelscited by 379
  35. 2020What is being transferred in transfer learning?cited by 368
  36. 2018Fairness Without Demographics in Repeated Loss Minimizationcited by 364
  37. 2020The Pitfalls of Simplicity Bias in Neural Networkscited by 351
  38. 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 343

50 most cited of 100 All 100 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 37; 2023: 78; 2024: 110; 2025: 90; 2026: 66.

381 papers.

Sort:

By area:

  1. 2023Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervisioncited by 467
  2. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 435
  3. 2023Where are we in the search for an Artificial Visual Cortex for Embodied Intelligence?cited by 218
  4. 2022Surgical Fine-Tuning Improves Adaptation to Distribution Shiftscited by 205
  5. 2023LIV: Language-Image Representations and Rewards for Robotic Controlcited by 188
  6. 2023Debiasing Vision-Language Models via Biased Promptscited by 149
  7. 2022Diverse Weight Averaging for Out-of-Distribution Generalizationcited by 148
  8. 2024WARM: On the Benefits of Weight Averaged Reward Modelscited by 147
  9. 2022On Feature Learning in the Presence of Spurious Correlationscited by 145
  10. 2022A Whac-A-Mole Dilemma: Shortcuts Come in Multiples Where Mitigating One Amplifies Otherscited by 85
  11. 2023Simple and Fast Group Robustness by Automatic Feature Reweightingcited by 68
  12. 2023On the Foundations of Shortcut Learningcited by 65
38 more
  1. 2024A Multimodal Automated Interpretability Agentcited by 63
  2. 2022Contrastive Adapters for Foundation Model Group Robustnesscited by 62
  3. 2024Position: Bayesian Deep Learning is Needed in the Age of Large-Scale AIcited by 62
  4. 2023Towards Last-layer Retraining for Group Robustness with Fewer Annotationscited by 54
  5. 2022Mechanistic Mode Connectivitycited by 50
  6. 2023AdaNPC: Exploring Non-Parametric Classifier for Test-Time Adaptationcited by 48
  7. 2024TuneTables: Context Optimization for Scalable Prior-Data Fitted Networkscited by 47
  8. 2023Identifying Spurious Biases Early in Training through the Lens of Simplicity Biascited by 44
  9. 2023Spawrious: A Benchmark for Fine Control of Spurious Correlation Biasescited by 41
  10. 2024The Clever Hans Mirage: A Comprehensive Survey on Spurious Correlations in Machine Learningcited by 41
  11. 2023Robust Learning with Progressive Data Expansion Against Spurious Correlationcited by 39
  12. 2023Understanding and Improving Feature Learning for Out-of-Distribution Generalizationcited by 35
  13. 2024CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?cited by 35
  14. 2024Overcoming Reward Overoptimization via Adversarial Policy Optimization with Lightweight Uncertainty Estimationcited by 33
  15. 2024MM-SpuBench: Towards Better Understanding of Spurious Biases in Multimodal LLMscited by 33
  16. 2024TabReD: Analyzing Pitfalls and Filling the Gaps in Tabular Deep Learning Benchmarkscited by 33
  17. 2022Avoiding spurious correlations via logit correctioncited by 32
  18. 2023Overwriting Pretrained Bias with Finetuning Datacited by 32
  19. 2023Leveraging sparse and shared feature activations for disentangled representation learningcited by 32
  20. 2022Unraveling Attention via Convex Duality: Analysis and Interpretations of Vision Transformerscited by 30
  21. 2024Calibrating Multi-modal Representations: A Pursuit of Group Robustness without Annotationscited by 30
  22. 2023Quantification of Uncertainty with Adversarial Modelscited by 28
  23. 2022Guillotine Regularization: Why removing layers is needed to improve generalization in Self-Supervised Learningcited by 26
  24. 2022Predicting is not Understanding: Recognizing and Addressing Underspecification in Machine Learningcited by 26
  25. 2023Discovering environments with XRMcited by 26
  26. 2023Spurious Feature Diversification Improves Out-of-distribution Generalizationcited by 26
  27. 2022Disentangled Explanations of Neural Network Predictions by Finding Relevant Subspacescited by 25
  28. 2023Group Robust Classification Without Any Group Informationcited by 25
  29. 2024FairerCLIP: Debiasing CLIP's Zero-Shot Predictions using Functions in RKHSscited by 25
  30. 2024Navigating Shortcuts, Spurious Correlations, and Confounders: From Origins via Detection to Mitigationcited by 25
  31. 2022How Robust is Unsupervised Representation Learning to Distribution Shift?cited by 24
  32. 2023Zero-Shot Robustification of Zero-Shot Modelscited by 24
  33. 2024The Clever Hans Effect in Unsupervised Learningcited by 24
  34. 2022Why does Throwing Away Data Improve Worst-Group Error?cited by 22
  35. 2022Freeze then Train: Towards Provable Representation Learning under Spurious Correlations and Feature Noisecited by 22
  36. 2024Understanding Linear Probing then Fine-tuning Language Models from NTK Perspectivecited by 22
  37. 2023What Can We Learn from Unlearnable Datasets?cited by 21
  38. 2023Spuriosity Didn't Kill the Classifier: Using Invariant Predictions to Harness Spurious Featurescited by 21

50 most cited of 381 All 381 (JSON)