mlsanityData to 7 Oct 2026

Understanding and Testing Generalization of Deep Networks on Out-of-Distribution Data

arXiv 2111.09190 ↗First posted 17 Nov 2021cs.LG cs.AIcited by 5

Not on the map. Why?

Abstract (arXiv)

Deep network models perform excellently on In-Distribution (ID) data, but can significantly fail on Out-Of-Distribution (OOD) data. While developing methods focus on improving OOD generalization, few attention has been paid to evaluating the capability of models to handle OOD data. This study is devoted to analyzing the problem of experimental ID test and designing OOD test paradigm to accurately evaluate the practical performance. Our analysis is based on an introduced categorization of three types of distribution shifts to generate OOD data. Main observations include: (1) ID test fails in neither reflecting the actual performance of a single model nor comparing between different models under OOD data. (2) The ID test failure can be ascribed to the learned marginal and conditional spurious correlations resulted from the corresponding distribution shifts. Based on this, we propose novel OOD test paradigms to evaluate the generalization capacity of models to unseen data, and discuss how to use OOD test results to find bugs of models to guide model debugging.

Key facts. Cited by 5 papers in the map (data through 7 Oct 2026). Builds on 38 of its 53 references that are in the map. First posted on arXiv 17 Nov 2021.

Builds on

38 of 53 references.

References · 38

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,267
  3. 2016Densely Connected Convolutional Networkscited by 6,691
  4. 2017MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applicationscited by 3,827
  5. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  6. 2015Domain-Adversarial Training of Neural Networkscited by 3,023
  7. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,857
  8. 2016Aggregated Residual Transformations for Deep Neural Networkscited by 2,540
  9. 2014Unsupervised Domain Adaptation by Backpropagationcited by 1,906
  10. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,842
  11. 2019Invariant Risk Minimizationcited by 1,832
  12. 2018Know What You Don't Know: Unanswerable Questions for SQuADcited by 1,538
26 more
  1. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
  2. 2019Natural Adversarial Examplescited by 1,401
  3. 2016SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model sizecited by 1,054
  4. 2020In Search of Lost Domain Generalizationcited by 1,042
  5. 2017Deeper, Broader and Artier Domain Generalizationcited by 983
  6. 2017The Implicit Bias of Gradient Descent on Separable Datacited by 649
  7. 2015Causal inference using invariant prediction: identification and confidence intervalscited by 533
  8. 2020Measuring Robustness to Natural Distribution Shifts in Image Classificationcited by 452
  9. 2018Generalizing to Unseen Domains via Adversarial Data Augmentationcited by 451
  10. 2017Don't Just Assume; Look and Answer: Overcoming Priors for Visual Question Answeringcited by 337
  11. 2017Unified Deep Supervised Domain Adaptation and Generalizationcited by 322
  12. 2020Pretrained Transformers Improve Out-of-Distribution Robustnesscited by 293
  13. 2018Generalizing Across Domains via Cross-Gradient Trainingcited by 268
  14. 2015Domain Generalization for Object Recognition with Multi-task Autoencoderscited by 253
  15. 2019Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigationcited by 252
  16. 2020Deep Domain-Adversarial Image Generation for Domain Generalisationcited by 231
  17. 2021Deep Stable Learning for Out-Of-Distribution Generalizationcited by 153
  18. 2020Understanding the Failure Modes of Out-of-Distribution Generalizationcited by 146
  19. 2020A Unified View of Label Shift Estimationcited by 119
  20. 2020The Effect of Natural Distribution Shift on Question Answering Modelscited by 104
  21. 2020On the Value of Out-of-Distribution Testing: An Example of Goodhart's Lawcited by 74
  22. 2019Stable Learning via Sample Reweightingcited by 70
  23. 2017Invariant Causal Prediction for Sequential Datacited by 63
  24. 2020Words aren't enough, their order matters: On the Robustness of Grounding Visual Referring Expressionscited by 33
  25. 2019Quantifying and Alleviating the Language Prior Problem in Visual Question Answeringcited by 13
  26. 2020Unseen Target Stance Detection with Adversarial Domain Generalizationcited by 11

Built on by

Cited by year

Citing papers per year: 2021: 1; 2022: 3; 2023: 1.

5 papers.

  1. 2022Towards out of distribution generalization for problems in mechanicscited by 11
  2. 2023Holistic chemical evaluation reveals pitfalls in reaction prediction modelscited by 4
  3. 2022Characterizing and Understanding the Behavior of Quantized Models for Reliable Deploymentcited by 2
  4. 2022LaF: Labeling-Free Model Selection for Automated Deep Neural Network Reusingcited by 1
  5. 2021A benchmark with decomposed distribution shifts for 360 monocular depth estimationcited by 0