mlsanityData to 7 Oct 2026

Just Train Twice: Improving Group Robustness without Training Group Information

arXiv 2107.09044 ↗First posted 19 Jul 2021cs.LG cs.AI cs.CYcited by 522

Not on the map. Why?

Abstract (arXiv)

Standard training via empirical risk minimization (ERM) can produce models that achieve high accuracy on average but low accuracy on certain groups, especially in the presence of spurious correlations between the input and label. Prior approaches that achieve high worst-group accuracy, like group distributionally robust optimization (group DRO) require expensive group annotations for each training point, whereas approaches that do not use such group annotations typically achieve unsatisfactory worst-group accuracy. In this paper, we propose a simple two-stage approach, JTT, that first trains a standard ERM model for several epochs, and then trains a second model that upweights the training examples that the first model misclassified. Intuitively, this upweights examples from groups on which standard ERM models perform poorly, leading to improved worst-group performance. Averaged over four image classification and natural language processing tasks with spurious correlations, JTT closes 75% of the gap in worst-group accuracy between standard ERM and group DRO, while only requiring group annotations on a small validation set in order to tune hyperparameters.

Key facts. Cited by 522 papers in the map (data through 7 Oct 2026). Builds on 43 of its 63 references that are in the map. First posted on arXiv 19 Jul 2021.

Builds on

43 of 63 references.

References · 43

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  3. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  4. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  5. 2013Intriguing properties of neural networkscited by 5,114
  6. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
  7. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  8. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,665
  9. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,201
  10. 2016Equality of Opportunity in Supervised Learningcited by 2,000
  11. 2016Understanding deep learning requires rethinking generalizationcited by 1,657
  12. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,366
31 more
  1. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,232
  2. 2018Generalized Cross Entropy Loss for Training Deep Neural Networks with Noisy Labelscited by 994
  3. 2019Learning Imbalanced Datasets with Label-Distribution-Aware Margin Losscited by 987
  4. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 864
  5. 2017Evasion Attacks against Machine Learning at Test Timecited by 686
  6. 2018Learning to Reweight Examples for Robust Deep Learningcited by 670
  7. 2018Annotation Artifacts in Natural Language Inference Datacited by 656
  8. 2013Domain Generalization via Invariant Feature Representationcited by 622
  9. 2018A Reductions Approach to Fair Classificationcited by 585
  10. 2019Agnostic Federated Learningcited by 451
  11. 2017On Fairness and Calibrationcited by 406
  12. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 396
  13. 2018Fairness Without Demographics in Repeated Loss Minimizationcited by 363
  14. 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
  15. 2020Environment Inference for Invariant Learningcited by 306
  16. 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 262
  17. 2018What is the Effect of Importance Weighting in Deep Learning?cited by 250
  18. 2018Multiaccuracy: Black-Box Post-Processing for Fairness in Classificationcited by 221
  19. 2019Hidden Stratification Causes Clinically Meaningful Failures in Machine Learning for Medical Imagingcited by 215
  20. 2020No Subclass Left Behind: Fine-Grained Robustness in Coarse-Grained Classification Problemscited by 208
  21. 2016Demographic Dialectal Variation in Social Media: A Case Study of African-American Englishcited by 194
  22. 2016Statistics of Robust Optimization: A Generalized Empirical Likelihood Approachcited by 182
  23. 2020Large-Scale Methods for Distributionally Robust Optimizationcited by 179
  24. 2019Distributionally Robust Language Modelingcited by 142
  25. 2020Learning from Failure: Training Debiased Classifier from Biased Classifiercited by 133
  26. 2020Towards Debiasing NLU Models from Unknown Biasescited by 116
  27. 2020Model Patching: Closing the Subgroup Performance Gap with Data Augmentationcited by 82
  28. 2020Coping with Label Shift via Distributionally Robust Optimisationcited by 69
  29. 2018Deep Learning Predicts Hip Fracture using Confounding Patient and Healthcare Variablescited by 50
  30. 2020Heteroskedastic and Imbalanced Deep Learning with Adaptive Regularizationcited by 49
  31. 2019Maximum Weighted Loss Discrepancycited by 11

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 18; 2022: 94; 2023: 119; 2024: 118; 2025: 103; 2026: 70.

522 papers.

Sort:

By area:

  1. 2023Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervisioncited by 445
  2. 2021Towards Out-Of-Distribution Generalization: A Surveycited by 431
  3. 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
  4. 2022Last Layer Re-Training is Sufficient for Robustness to Spurious Correlationscited by 379
  5. 2022Surgical Fine-Tuning Improves Adaptation to Distribution Shiftscited by 205
  6. 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
  7. 2021Simple data balancing achieves competitive worst-group-accuracycited by 177
  8. 2021A Fine-Grained Analysis on Distribution Shiftcited by 171
  9. 2022Correct-N-Contrast: A Contrastive Approach for Improving Robustness to Spurious Correlationscited by 171
  10. 2022DeepCore: A Comprehensive Library for Coreset Selection in Deep Learningcited by 146
  11. 2022On Feature Learning in the Presence of Spurious Correlationscited by 145
  12. 2023Change is Hard: A Closer Look at Subpopulation Shiftcited by 119
38 more
  1. 2021Measure and Improve Robustness in NLP Models: A Surveycited by 107
  2. 2024Entropy is not Enough for Test-Time Adaptation: From the Perspective of Disentangled Factorscited by 106
  3. 2022Test-Time Adaptation via Conjugate Pseudo-labelscited by 98
  4. 2022Spread Spurious Attribute: Improving Worst-group Accuracy with Spurious Attribute Estimationcited by 88
  5. 2022A Whac-A-Mole Dilemma: Shortcuts Come in Multiples Where Mitigating One Amplifies Otherscited by 85
  6. 2022Shortcut Learning of Large Language Models in Natural Language Understandingcited by 83
  7. 2022Distilling Model Failures as Directions in Latent Spacecited by 82
  8. 2022Perceptual Grouping in Contrastive Vision-Language Modelscited by 82
  9. 2023Discover and Cure: Concept-aware Mitigation of Spurious Correlationcited by 71
  10. 2022Agree to Disagree: Diversity through Disagreement for Better Transferabilitycited by 69
  11. 2023Simple and Fast Group Robustness by Automatic Feature Reweightingcited by 68
  12. 2022MaskTune: Mitigating Spurious Correlations by Forcing to Explorecited by 66
  13. 2022Contrastive Adapters for Foundation Model Group Robustnesscited by 62
  14. 2023Towards Coherent Image Inpainting Using Denoising Diffusion Implicit Modelscited by 61
  15. 2022ZIN: When and How to Learn Invariance Without Environment Partition?cited by 58
  16. 2023Discovering and Mitigating Visual Biases through Keyword Explanationcited by 58
  17. 2023Diagnosing and Rectifying Vision Models using Languagecited by 58
  18. 2021Detecting Corrupted Labels Without Training a Model to Predictcited by 57
  19. 2022Let Invariant Rationale Discovery Inspire Graph Contrastive Learningcited by 57
  20. 2022Diversify and Disambiguate: Learning From Underspecified Datacited by 55
  21. 2023Towards Last-layer Retraining for Group Robustness with Fewer Annotationscited by 54
  22. 2022Improving the Fairness of Chest X-ray Classifierscited by 52
  23. 2022Detecting Shortcut Learning for Fair Medical AI using Shortcut Testingcited by 52
  24. 2022Learning Debiased Classifier with Biased Committeecited by 51
  25. 2023Explore Spurious Correlations at the Concept Level in Language Models for Text Classificationcited by 48
  26. 2023Mitigating Spurious Correlations in Multi-modal Models during Fine-tuningcited by 47
  27. 2023Identifying Spurious Biases Early in Training through the Lens of Simplicity Biascited by 44
  28. 2023Model Agnostic Sample Reweighting for Out-of-Distribution Learningcited by 43
  29. 2024LLM as Dataset Analyst: Subpopulation Structure Discovery with Large Language Modelcited by 43
  30. 2023Does Invariant Graph Learning via Environment Augmentation Learn Invariance?cited by 42
  31. 2022The Road to Explainability is Paved with Bias: Measuring the Fairness of Explanationscited by 41
  32. 2023Spawrious: A Benchmark for Fine Control of Spurious Correlation Biasescited by 41
  33. 2022Perfectly Balanced: Improving Transfer and Robustness of Supervised Contrastive Learningcited by 40
  34. 2023Robust Learning with Progressive Data Expansion Against Spurious Correlationcited by 39
  35. 2024The Clever Hans Mirage: A Comprehensive Survey on Spurious Correlations in Machine Learningcited by 39
  36. 2022Pareto Invariant Risk Minimization: Towards Mitigating the Optimization Dilemma in Out-of-Distribution Generalizationcited by 38
  37. 2022Rich Feature Construction for the Optimization-Generalization Dilemmacited by 34
  38. 2022Calibrated ensembles can mitigate accuracy tradeoffs under distribution shiftcited by 34

50 most cited of 522 All 522 (JSON)