mlsanityData to 14 Oct 2026

Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimation

arXiv 2310.11991 ↗First posted 18 Oct 2023cs.LG stat.MLcited by 6

Not on the map. Why?

Abstract (arXiv)

Out-of-distribution generalization in neural networks is often hampered by spurious correlations. A common strategy is to mitigate this by removing spurious concepts from the neural network representation of the data. Existing concept-removal methods tend to be overzealous by inadvertently eliminating features associated with the main task of the model, thereby harming model performance. We propose an iterative algorithm that separates spurious from main-task concepts by jointly identifying two low-dimensional orthogonal subspaces in the neural network representation. We evaluate the algorithm on benchmark datasets for computer vision (Waterbirds, CelebA) and natural language processing (MultiNLI), and show that it outperforms existing concept removal methods

Key facts. Cited by 6 papers in the map (data through 14 Oct 2026). Builds on 56 of its 71 references that are in the map. First posted on arXiv 18 Oct 2023.

Builds on

56 of 71 references.

References · 56

  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. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,733
  4. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  5. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,682
  6. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,218
  7. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  8. 2014Unsupervised Domain Adaptation by Backpropagationcited by 1,917
  9. 2019Invariant Risk Minimizationcited by 1,854
  10. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,432
  11. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  12. 2018On the Spectral Bias of Neural Networkscited by 1,341
38 more
  1. 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,319
  2. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,312
  3. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,247
  4. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  5. 2021Probing Classifiers: Promises, Shortcomings, and Advancescited by 876
  6. 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 708
  7. 2018Annotation Artifacts in Natural Language Inference Datacited by 671
  8. 2018Mitigating Unwanted Biases with Adversarial Learningcited by 663
  9. 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 523
  10. 2020Underspecification Presents Challenges for Credibility in Modern Machine Learningcited by 496
  11. 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 455
  12. 2022Last Layer Re-Training is Sufficient for Robustness to Spurious Correlationscited by 381
  13. 2021Interpretable Machine Learning: Fundamental Principles and 10 Grand Challengescited by 367
  14. 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 343
  15. 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
  16. 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 267
  17. 2023LEACE: Perfect linear concept erasure in closed formcited by 267
  18. 2020Concept Whitening for Interpretable Image Recognitioncited by 225
  19. 2021Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalizationcited by 220
  20. 2019Probing Neural Network Comprehension of Natural Language Argumentscited by 214
  21. 2015Censoring Representations with an Adversarycited by 192
  22. 2021Simple data balancing achieves competitive worst-group-accuracycited by 178
  23. 2018Adversarial Removal of Demographic Attributes from Text Datacited by 170
  24. 2020Probing the Probing Paradigm: Does Probing Accuracy Entail Task Relevance?cited by 151
  25. 2022On Feature Learning in the Presence of Spurious Correlationscited by 145
  26. 2021Salient ImageNet: How to discover spurious features in Deep Learning?cited by 114
  27. 2022Linear Adversarial Concept Erasurecited by 101
  28. 2018How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarkscited by 88
  29. 2016Places: An Image Database for Deep Scene Understandingcited by 76
  30. 2022Domain-Adjusted Regression or: ERM May Already Learn Features Sufficient for Out-of-Distribution Generalizationcited by 69
  31. 2021Examining and Combating Spurious Features under Distribution Shiftcited by 65
  32. 2019Attenuating Bias in Word Vectorscited by 64
  33. 2020Robustness to Spurious Correlations via Human Annotationscited by 62
  34. 2019Understanding Undesirable Word Embedding Associationscited by 59
  35. 2020Identifying Spurious Correlations for Robust Text Classificationcited by 59
  36. 2021Counterfactual Interventions Reveal the Causal Effect of Relative Clause Representations on Agreement Predictioncited by 59
  37. 2022Discovering Latent Concepts Learned in BERTcited by 56
  38. 2022Kernelized Concept Erasurecited by 48

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 3; 2025: 1; 2026: 2.

6 papers.

  1. 2024The Clever Hans Mirage: A Comprehensive Survey on Spurious Correlations in Machine Learningcited by 41
  2. 2024Navigating Shortcuts, Spurious Correlations, and Confounders: From Origins via Detection to Mitigationcited by 25
  3. 2025Preserving Task-Relevant Information Under Linear Concept Removalcited by 4
  4. 2024Optimizing importance weighting in the presence of sub-population shiftscited by 3
  5. 2026Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositionscited by 0
  6. 2026Operational Abstractions of Neural Network Concepts via Topological Representationscited by 0