Data to 7 Oct 2026

Understanding the Failure Modes of Out-of-Distribution Generalization

arXiv 2010.15775 ↗First posted 29 Oct 2020cs.LG cs.CV stat.MLcited by 146

Not on the map. Why?

Abstract (arXiv)

Empirical studies suggest that machine learning models often rely on features, such as the background, that may be spuriously correlated with the label only during training time, resulting in poor accuracy during test-time. In this work, we identify the fundamental factors that give rise to this behavior, by explaining why models fail this way em even in easy-to-learn tasks where one would expect these models to succeed. In particular, through a theoretical study of gradient-descent-trained linear classifiers on some easy-to-learn tasks, we uncover two complementary failure modes. These modes arise from how spurious correlations induce two kinds of skews in the data: one geometric in nature, and another, statistical in nature. Finally, we construct natural modifications of image classification datasets to understand when these failure modes can arise in practice. We also design experiments to isolate the two failure modes when training modern neural networks on these datasets.

Key facts. Cited by 146 papers in the map (data through 7 Oct 2026). First posted on arXiv 29 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 17; 2022: 46; 2023: 28; 2024: 26; 2025: 17; 2026: 12.

146 papers.

Sort:

By area:

2022Last Layer Re-Training is Sufficient for Robustness to Spurious Correlationscited by 379
2021Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalizationcited by 220
2022Exploring Length Generalization in Large Language Modelscited by 204
2022Out-of-Distribution Detection and Selective Generation for Conditional Language Modelscited by 193
2022Correct-N-Contrast: A Contrastive Approach for Improving Robustness to Spurious Correlationscited by 171
2022Handling Distribution Shifts on Graphs: An Invariance Perspectivecited by 163
2024Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelscited by 158
2022Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphscited by 144
2022Leveraging Unlabeled Data to Predict Out-of-Distribution Performancecited by 141
2021Deep Learning Through the Lens of Example Difficultycited by 139
2021Does Invariant Risk Minimization Capture Invariance?cited by 116
2021Model-Based Domain Generalizationcited by 103

38 more

2021Towards a Theoretical Framework of Out-of-Distribution Generalizationcited by 94
2023Eliciting Human Preferences with Language Modelscited by 90
2021Invariant Information Bottleneck for Domain Generalizationcited by 86
2022Sparse Mixture-of-Experts are Domain Generalizable Learnerscited by 85
2024Transformers Can Achieve Length Generalization But Not Robustlycited by 82
2021Can Subnetwork Structure be the Key to Out-of-Distribution Generalization?cited by 75
2021Prior and Posterior Networks: A Survey on Evidential Deep Learning Methods For Uncertainty Estimationcited by 71
2022Probable Domain Generalization via Quantile Risk Minimizationcited by 66
2023On the Foundations of Shortcut Learningcited by 64
2022Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlationcited by 59
2021Causally motivated Shortcut Removal Using Auxiliary Labelscited by 55
2023Identifying Spurious Biases Early in Training through the Lens of Simplicity Biascited by 44
2021Nonlinear Invariant Risk Minimization: A Causal Approachcited by 42
2025Beyond Reward Hacking: Causal Rewards for Large Language Model Alignmentcited by 42
2023Spawrious: A Benchmark for Fine Control of Spurious Correlation Biasescited by 41
2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 40
2021Finding and Fixing Spurious Patterns with Explanationscited by 38
2022Calibrated ensembles can mitigate accuracy tradeoffs under distribution shiftcited by 34
2022Active Learning Helps Pretrained Models Learn the Intended Taskcited by 33
2022Avoiding spurious correlations via logit correctioncited by 32
2021Dangers of Bayesian Model Averaging under Covariate Shiftcited by 31
2023Don't blame Dataset Shift! Shortcut Learning due to Gradients and Cross Entropycited by 30
2023Overcoming Simplicity Bias in Deep Networks using a Feature Sievecited by 29
2022Improving Multi-Task Generalization via Regularizing Spurious Correlationcited by 28
2022Understanding Rare Spurious Correlations in Neural Networkscited by 27
2022Out-of-distribution Detection via Frequency-regularized Generative Modelscited by 23
2022Why does Throwing Away Data Improve Worst-Group Error?cited by 22
2024Navigating Shortcuts, Spurious Correlations, and Confounders: From Origins via Detection to Mitigationcited by 22
2023Spuriosity Didn't Kill the Classifier: Using Invariant Predictions to Harness Spurious Featurescited by 21
2025Towards LLMs Robustness to Changes in Prompt Format Stylescited by 21
2022Igeood: An Information Geometry Approach to Out-of-Distribution Detectioncited by 20
2022Explaining the physics of transfer learning a data-driven subgrid-scale closure to a different turbulent flowcited by 20
2024When does compositional structure yield compositional generalization? A kernel theorycited by 19
2022Nuisances via Negativa: Adjusting for Spurious Correlations via Data Augmentationcited by 18
2022Improving Out-of-Distribution Generalization by Adversarial Training with Structured Priorscited by 17
2022Test-Time Training for Graph Neural Networkscited by 17
2022Bayesian Active Learning for Scanning Probe Microscopy: from Gaussian Processes to Hypothesis Learningcited by 16
2022Training Debiased Subnetworks with Contrastive Weight Pruningcited by 16

50 most cited of 146 All 146 (JSON)