mlsanityData to 7 Oct 2026

Removing Spurious Features can Hurt Accuracy and Affect Groups Disproportionately

arXiv 2012.04104 ↗First posted 7 Dec 2020cs.LG cs.AI cs.CYcited by 52

Not on the map. Why?

Abstract (arXiv)

The presence of spurious features interferes with the goal of obtaining robust models that perform well across many groups within the population. A natural remedy is to remove spurious features from the model. However, in this work we show that removal of spurious features can decrease accuracy due to the inductive biases of overparameterized models. We completely characterize how the removal of spurious features affects accuracy across different groups (more generally, test distributions) in noiseless overparameterized linear regression. In addition, we show that removal of spurious feature can decrease the accuracy even in balanced datasets -- each target co-occurs equally with each spurious feature; and it can inadvertently make the model more susceptible to other spurious features. Finally, we show that robust self-training can remove spurious features without affecting the overall accuracy. Experiments on the Toxic-Comment-Detectoin and CelebA datasets show that our results hold in non-linear models.

Key facts. Cited by 52 papers in the map (data through 7 Oct 2026). First posted on arXiv 7 Dec 2020.

Builds on

No reference list yet.

Built on by

Cited by year

Citing papers per year: 2021: 11; 2022: 22; 2023: 11; 2024: 6; 2025: 2.

52 papers.

Sort:

By area:

  1. 2022Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distributioncited by 722
  2. 2021Data Collection and Quality Challenges in Deep Learning: A Data-Centric AI Perspectivecited by 116
  3. 2021Salient ImageNet: How to discover spurious features in Deep Learning?cited by 114
  4. 2022Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?cited by 89
  5. 2021Identifying and Mitigating Spurious Correlations for Improving Robustness in NLP Modelscited by 76
  6. 2021Can Subnetwork Structure be the Key to Out-of-Distribution Generalization?cited by 75
  7. 2023Discover and Cure: Concept-aware Mitigation of Spurious Correlationcited by 71
  8. 2022MaskTune: Mitigating Spurious Correlations by Forcing to Explorecited by 66
  9. 2022Post hoc Explanations may be Ineffective for Detecting Unknown Spurious Correlationcited by 59
  10. 2022A Comprehensive Study of Image Classification Model Sensitivity to Foregrounds, Backgrounds, and Visual Attributescited by 56
  11. 2021Sample Selection for Fair and Robust Trainingcited by 48
  12. 2021Understanding and Improving Robustness of Vision Transformers through Patch-based Negative Augmentationcited by 40
38 more
  1. 2024InjecGuard: Benchmarking and Mitigating Over-defense in Prompt Injection Guardrail Modelscited by 36
  2. 2022Calibrated ensembles can mitigate accuracy tradeoffs under distribution shiftcited by 34
  3. 2021An Empirical Framework for Domain Generalization in Clinical Settingscited by 32
  4. 2022Improving Multi-Task Generalization via Regularizing Spurious Correlationcited by 28
  5. 2022Understanding Rare Spurious Correlations in Neural Networkscited by 27
  6. 2023People Make Better Edits: Measuring the Efficacy of LLM-Generated Counterfactually Augmented Data for Harmful Language Detectioncited by 25
  7. 2022Explicit Tradeoffs between Adversarial and Natural Distributional Robustnesscited by 20
  8. 2021Understanding and Mitigating Accuracy Disparity in Regressioncited by 17
  9. 2023Causality and Independence Enhancement for Biased Node Classificationcited by 13
  10. 2022A Sandbox Tool to Bias(Stress)-Test Fairness Algorithmscited by 12
  11. 2022Counterfactually Augmented Data and Unintended Bias: The Case of Sexism and Hate Speech Detectioncited by 11
  12. 2022Simplicity Bias Leads to Amplified Performance Disparitiescited by 11
  13. 2022Task Discovery: Finding the Tasks that Neural Networks Generalize oncited by 9
  14. 2023Preemptively Pruning Clever-Hans Strategies in Deep Neural Networkscited by 9
  15. 2024Understanding Domain Generalization: A Noise Robustness Perspectivecited by 8
  16. 2023Toward Operationalizing Pipeline-aware ML Fairness: A Research Agenda for Developing Practical Guidelines and Toolscited by 7
  17. 2022CLAD: A Contrastive Learning based Approach for Background Debiasingcited by 6
  18. 2023Spurious Correlations and Where to Find Themcited by 6
  19. 2023The Pursuit of Human Labeling: A New Perspective on Unsupervised Learningcited by 6
  20. 2024Robust Fine-tuning of Zero-shot Models via Variance Reductioncited by 6
  21. 2022OOD-Probe: A Neural Interpretation of Out-of-Domain Generalizationcited by 5
  22. 2023Fairness and Sequential Decision Making: Limits, Lessons, and Opportunitiescited by 5
  23. 2023ASPIRE: Language-Guided Data Augmentation for Improving Robustness Against Spurious Correlationscited by 5
  24. 2021The Effect of Model Size on Worst-Group Generalizationcited by 4
  25. 2022PatchMix Augmentation to Identify Causal Features in Few-shot Learningcited by 4
  26. 2024LEVI: Generalizable Fine-tuning via Layer-wise Ensemble of Different Viewscited by 4
  27. 2021Human Imperceptible Attacks and Applications to Improve Fairnesscited by 3
  28. 2022Conditional entropy minimization principle for learning domain invariant representation featurescited by 3
  29. 2022Unveiling the Tapestry: the Interplay of Generalization and Forgetting in Continual Learningcited by 3
  30. 2025Improving Group Robustness on Spurious Correlation via Evidential Alignmentcited by 3
  31. 2022Learning to Increase the Power of Conditional Randomization Testscited by 2
  32. 2022How Much User Context Do We Need? Privacy by Design in Mental Health NLP Applicationcited by 2
  33. 2023Collaborative Development of NLP modelscited by 2
  34. 2024Formalising Anti-Discrimination Law in Automated Decision Systemscited by 2
  35. 2024Towards Utilising a Range of Neural Activations for Comprehending Representational Associationscited by 2
  36. 2022On Understanding the Influence of Controllable Factors with a Feature Attribution Algorithm: a Medical Case Studycited by 1
  37. 2023Degraded Polygons Raise Fundamental Questions of Neural Network Perceptioncited by 1
  38. 2021On the Importance of Regularisation & Auxiliary Information in OOD Detectioncited by 0

50 most cited of 52 All 52 (JSON)