Data to 7 Oct 2026

An Investigation of Why Overparameterization Exacerbates Spurious Correlations

arXiv 2005.04345 ↗First posted 9 May 2020cs.LG cs.CV stat.MLcited by 310

Not on the map. Why?

Abstract (arXiv)

We study why overparameterization -- increasing model size well beyond the point of zero training error -- can hurt test error on minority groups despite improving average test error when there are spurious correlations in the data. Through simulations and experiments on two image datasets, we identify two key properties of the training data that drive this behavior: the proportions of majority versus minority groups, and the signal-to-noise ratio of the spurious correlations. We then analyze a linear setting and theoretically show how the inductive bias of models towards "memorizing" fewer examples can cause overparameterization to hurt. Our analysis leads to a counterintuitive approach of subsampling the majority group, which empirically achieves low minority error in the overparameterized regime, even though the standard approach of upweighting the minority fails. Overall, our results suggest a tension between using overparameterized models versus using all the training data for achieving low worst-group error.

Key facts. Cited by 310 papers in the map (data through 7 Oct 2026). First posted on arXiv 9 May 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 43; 2022: 84; 2023: 58; 2024: 61; 2025: 39; 2026: 25.

310 papers.

Sort:

By area:

2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 522
2021Towards Out-Of-Distribution Generalization: A Surveycited by 431
2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
2021Fishr: Invariant Gradient Variances for Out-of-Distribution Generalizationcited by 186
2022Datamodels: Predicting Predictions from Training Datacited by 186
2021Simple data balancing achieves competitive worst-group-accuracycited by 177
2022On Feature Learning in the Presence of Spurious Correlationscited by 145
2021Adapting Language Models for Zero-shot Learning by Meta-tuning on Dataset and Prompt Collectionscited by 128
2021Do We Actually Need Dense Over-Parameterization? In-Time Over-Parameterization in Sparse Trainingcited by 107
2021Measure and Improve Robustness in NLP Models: A Surveycited by 107
2021Heterogeneous Risk Minimizationcited by 90
2023Eliciting Human Preferences with Language Modelscited by 90

38 more

2021Evading the Simplicity Bias: Training a Diverse Set of Models Discovers Solutions with Superior OOD Generalizationcited by 83
2024Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Modelscited by 78
2023A Theoretical Understanding of Shallow Vision Transformers: Learning, Generalization, and Sample Complexitycited by 77
2021Identifying and Mitigating Spurious Correlations for Improving Robustness in NLP Modelscited by 76
2021Can Subnetwork Structure be the Key to Out-of-Distribution Generalization?cited by 75
2021Label-Imbalanced and Group-Sensitive Classification under Overparameterizationcited by 74
2022GLUE-X: Evaluating Natural Language Understanding Models from an Out-of-distribution Generalization Perspectivecited by 72
2023Fast Model Debias with Machine Unlearningcited by 70
2021Examining and Combating Spurious Features under Distribution Shiftcited by 64
2023On the Foundations of Shortcut Learningcited by 64
2022Contrastive Adapters for Foundation Model Group Robustnesscited by 62
2022Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlationcited by 59
2021Causally motivated Shortcut Removal Using Auxiliary Labelscited by 55
2021DORO: Distributional and Outlier Robust Optimizationcited by 54
2022DrugOOD: Out-of-Distribution (OOD) Dataset Curator and Benchmark for AI-aided Drug Discovery -- A Focus on Affinity Prediction Problems with Noise Annotationscited by 54
2022Mechanistic Mode Connectivitycited by 50
2023Measuring Inductive Biases of In-Context Learning with Underspecified Demonstrationscited by 49
2023Explore Spurious Correlations at the Concept Level in Language Models for Text Classificationcited by 48
2021Unsupervised Learning of Debiased Representations with Pseudo-Attributescited by 44
2023Identifying Spurious Biases Early in Training through the Lens of Simplicity Biascited by 44
2024Challenging Forgets: Unveiling the Worst-Case Forget Sets in Machine Unlearningcited by 44
2023Model Agnostic Sample Reweighting for Out-of-Distribution Learningcited by 43
2021Out-of-distribution Generalization in the Presence of Nuisance-Induced Spurious Correlationscited by 41
2023Spawrious: A Benchmark for Fine Control of Spurious Correlation Biasescited by 41
2021Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspectivecited by 40
2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 40
2023Robust and Interpretable Medical Image Classifiers via Concept Bottleneck Modelscited by 40
2022On-Demand Sampling: Learning Optimally from Multiple Distributionscited by 39
2023Robust Learning with Progressive Data Expansion Against Spurious Correlationcited by 39
2024The Clever Hans Mirage: A Comprehensive Survey on Spurious Correlations in Machine Learningcited by 39
2024A Sober Look at the Robustness of CLIPs to Spurious Featurescited by 38
2022Are Vision Transformers Robust to Spurious Correlations?cited by 36
2024CLIP the Bias: How Useful is Balancing Data in Multimodal Learning?cited by 35
2021Are Larger Pretrained Language Models Uniformly Better? Comparing Performance at the Instance Levelcited by 34
2022Calibrated ensembles can mitigate accuracy tradeoffs under distribution shiftcited by 34
2022UMIX: Improving Importance Weighting for Subpopulation Shift via Uncertainty-Aware Mixupcited by 34
2022Active Learning Helps Pretrained Models Learn the Intended Taskcited by 33
2024MM-SpuBench: Towards Better Understanding of Spurious Biases in Multimodal LLMscited by 33

50 most cited of 310 All 310 (JSON)