mlsanityData to 14 Oct 2026

Modeling the Data-Generating Process is Necessary for Out-of-Distribution Generalization

arXiv 2206.07837 ↗First posted 15 Jun 2022cs.LG cs.AIcited by 34

Not on the map. Why?

Abstract (arXiv)

Recent empirical studies on domain generalization (DG) have shown that DG algorithms that perform well on some distribution shifts fail on others, and no state-of-the-art DG algorithm performs consistently well on all shifts. Moreover, real-world data often has multiple distribution shifts over different attributes; hence we introduce multi-attribute distribution shift datasets and find that the accuracy of existing DG algorithms falls even further. To explain these results, we provide a formal characterization of generalization under multi-attribute shifts using a canonical causal graph. Based on the relationship between spurious attributes and the classification label, we obtain realizations of the canonical causal graph that characterize common distribution shifts and show that each shift entails different independence constraints over observed variables. As a result, we prove that any algorithm based on a single, fixed constraint cannot work well across all shifts, providing theoretical evidence for mixed empirical results on DG algorithms. Based on this insight, we develop Causally Adaptive Constraint Minimization (CACM), an algorithm that uses knowledge about the data-generating process to adaptively identify and apply the correct independence constraints for regularization. Results on fully synthetic, MNIST, small NORB, and Waterbirds datasets, covering binary and multi-valued attributes and labels, show that adaptive dataset-dependent constraints lead to the highest accuracy on unseen domains whereas incorrect constraints fail to do so. Our results demonstrate the importance of modeling the causal relationships inherent in the data-generating process.

Key facts. Cited by 34 papers in the map (data through 14 Oct 2026). Builds on 37 of its 63 references that are in the map. First posted on arXiv 15 Jun 2022.

Builds on

37 of 63 references.

References · 37

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2015Domain-Adversarial Training of Neural Networkscited by 3,044
  3. 2019RandAugment: Practical automated data augmentation with a reduced search spacecited by 2,269
  4. 2019Invariant Risk Minimizationcited by 1,854
  5. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,375
  6. 2016Deep CORAL: Correlation Alignment for Deep Domain Adaptationcited by 1,200
  7. 2020In Search of Lost Domain Generalizationcited by 1,054
  8. 2021Generalizing to Unseen Domains: A Survey on Domain Generalizationcited by 766
  9. 2021Domain Generalization: A Surveycited by 764
  10. 2017Learning to Generalize: Meta-Learning for Domain Generalizationcited by 763
  11. 2020Out-of-Distribution Generalization via Risk Extrapolation (REx)cited by 742
  12. 2013Domain Generalization via Invariant Feature Representationcited by 623
25 more
  1. 2020Self-Challenging Improves Cross-Domain Generalizationcited by 426
  2. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 404
  3. 2020The Pitfalls of Simplicity Bias in Neural Networkscited by 351
  4. 2017Functional Map of the Worldcited by 338
  5. 2020The Risks of Invariant Risk Minimizationcited by 262
  6. 2015Domain Generalization for Object Recognition with Multi-task Autoencoderscited by 253
  7. 2020Weakly-Supervised Disentanglement Without Compromisescited by 239
  8. 2020Domain Generalization using Causal Matchingcited by 225
  9. 2021Invariance Principle Meets Information Bottleneck for Out-of-Distribution Generalizationcited by 220
  10. 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
  11. 2020Improve Unsupervised Domain Adaptation with Mixup Trainingcited by 185
  12. 2019Quantifying the effects of data augmentation and stain color normalization in convolutional neural networks for computational pathologycited by 173
  13. 2021A Fine-Grained Analysis on Distribution Shiftcited by 171
  14. 2018Reducing Gender Bias in Abusive Language Detectioncited by 152
  15. 2018Domain Generalization via Conditional Invariant Representationcited by 146
  16. 2020Efficient Domain Generalization via Common-Specific Low-Rank Decompositioncited by 134
  17. 2019Generalizing to unseen domains via distribution matchingcited by 110
  18. 2015Scatter Component Analysis: A Unified Framework for Domain Adaptation and Domain Generalizationcited by 96
  19. 2019Support and Invertibility in Domain-Invariant Representationscited by 87
  20. 2021OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalizationcited by 86
  21. 2017Machine learning methods for histopathological image analysiscited by 79
  22. 2019Domain Generalization via Multidomain Discriminant Analysiscited by 69
  23. 2021Causally motivated Shortcut Removal Using Auxiliary Labelscited by 55
  24. 2019Adversarial Invariant Feature Learning with Accuracy Constraint for Domain Generalizationcited by 43
  25. 2019On Learning Invariant Representation for Domain Adaptationcited by 39

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 4; 2023: 9; 2024: 10; 2025: 6; 2026: 5.

34 papers.

Sort:

By area:

  1. 2023Compositional Abilities Emerge Multiplicatively: Exploring Diffusion Models on a Synthetic Taskcited by 93
  2. 2022Mechanistic Mode Connectivitycited by 50
  3. 2024Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Spacecited by 35
  4. 2023Joint Learning of Label and Environment Causal Independence for Graph Out-of-Distribution Generalizationcited by 29
  5. 2024A Survey on Evaluation of Out-of-Distribution Generalizationcited by 21
  6. 2022Domain Generalization -- A Causal Perspectivecited by 14
  7. 2023A Review of the Role of Causality in Developing Trustworthy AI Systemscited by 14
  8. 2023OOD-CV-v2: An extended Benchmark for Robustness to Out-of-Distribution Shifts of Individual Nuisances in Natural Imagescited by 12
  9. 2024LCA-on-the-Line: Benchmarking Out-of-Distribution Generalization with Class Taxonomiescited by 12
  10. 2024Do causal predictors generalize better to new domains?cited by 11
  11. 2023Emerging Synergies in Causality and Deep Generative Models: A Surveycited by 10
  12. 2024Mind the Graph When Balancing Data for Fairness or Robustnesscited by 9
22 more
  1. 2023Domain constraints improve risk prediction when outcome data is missingcited by 8
  2. 2022Beyond Invariance: Test-Time Label-Shift Adaptation for Distributions with "Spurious" Correlationscited by 7
  3. 2022Malign Overfitting: Interpolation Can Provably Preclude Invariancecited by 6
  4. 2023A Markov Random Field model for Hypergraph-based Machine Learningcited by 5
  5. 2024Scalable Out-of-distribution Robustness in the Presence of Unobserved Confounderscited by 5
  6. 2025Mitigating Spurious Correlations in LLMs via Causality-Aware Post-Trainingcited by 5
  7. 2024Causally Inspired Regularization Enables Domain General Representationscited by 4
  8. 2023Emergence and Causality in Complex Systems: A Survey on Causal Emergence and Related Quantitative Studiescited by 3
  9. 2025Do regularization methods for shortcut mitigation work as intended?cited by 3
  10. 2025Heterogeneous Graph Structure Learning through the Lens of Data-generating Processescited by 2
  11. 2025Test-Time Adaptation by Causal Trimmingcited by 2
  12. 2023Causal Effect Regularization: Automated Detection and Removal of Spurious Attributescited by 1
  13. 2024Domain Adaptation for Sustainable Soil Management using Causal and Contrastive Constraint Minimizationcited by 1
  14. 2024A Unified Causal View of Instruction Tuningcited by 1
  15. 2026Self-Supervised Learning from Structural Invariancecited by 1
  16. 2026Does Your Neural Network Extrapolate? Feature Engineering as Identifiability Bias for OOD Generalizationcited by 1
  17. 2026Structure-agnostic Causal Representation Learningcited by 1
  18. 2024Invariant Correlation of Representation with Label: Enhancing Domain Generalization in Noisy Environmentscited by 0
  19. 2025DISCO: Mitigating Bias in Deep Learning with Conditional Distance Correlationcited by 0
  20. 2025Evaluating and Learning Robust Bandit Policies Under Uncertain Causal Mechanismscited by 0
  21. 2026A Causal Framework for Mitigating Data Shifts in Healthcarecited by 0
  22. 2026Can Circuit Alignment Predict OOD Generalization?cited by 0