mlsanityData to 7 Oct 2026

Understanding deep learning requires rethinking generalization

arXiv 1611.03530 ↗First posted 10 Nov 2016cs.LGcited by 1,657

Not on the map. Why?

Cited by 33 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance. Conventional wisdom attributes small generalization error either to properties of the model family, or to the regularization techniques used during training. Through extensive systematic experiments, we show how these traditional approaches fail to explain why large neural networks generalize well in practice. Specifically, our experiments establish that state-of-the-art convolutional networks for image classification trained with stochastic gradient methods easily fit a random labeling of the training data. This phenomenon is qualitatively unaffected by explicit regularization, and occurs even if we replace the true images by completely unstructured random noise. We corroborate these experimental findings with a theoretical construction showing that simple depth two neural networks already have perfect finite sample expressivity as soon as the number of parameters exceeds the number of data points as it usually does in practice. We interpret our experimental findings by comparison with traditional models.

Key facts. Cited by 1,657 papers in the map (data through 7 Oct 2026). First posted on arXiv 10 Nov 2016.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2018: 1; 2019: 2; 2020: 2; 2021: 481; 2022: 324; 2023: 224; 2024: 197; 2025: 210; 2026: 216.

1,657 papers.

Sort:

By area:

  1. 2020GShard: Scaling Giant Models with Conditional Computation and Automatic Shardingcited by 1,601
  2. 2021An Explanation of In-context Learning as Implicit Bayesian Inferencecited by 915
  3. 2022Quantifying Memorization Across Neural Language Modelscited by 755
  4. 2022Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasetscited by 599
  5. 2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 529
  6. 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 522
  7. 2018Scalable agent alignment via reward modeling: a research directioncited by 438
  8. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 415
  9. 2021Large Language Models Can Be Strong Differentially Private Learnerscited by 405
  10. 2021Open-Set Recognition: a Good Closed-Set Classifier is All You Need?cited by 351
  11. 2021Gradient Projection Memory for Continual Learningcited by 338
  12. 2021Pre-Trained Models: Past, Present and Futurecited by 319
38 more
  1. 2021Membership Inference Attacks on Machine Learning: A Surveycited by 306
  2. 2021The Role of Permutation Invariance in Linear Mode Connectivity of Neural Networkscited by 285
  3. 2021A Geometric Analysis of Neural Collapse with Unconstrained Featurescited by 226
  4. 2022Memorization Without Overfitting: Analyzing the Training Dynamics of Large Language Modelscited by 218
  5. 2023Better Diffusion Models Further Improve Adversarial Trainingcited by 207
  6. 2021Trustworthy AI: From Principles to Practicescited by 206
  7. 2021Fixing Data Augmentation to Improve Adversarial Robustnesscited by 201
  8. 2023ReLoRA: High-Rank Training Through Low-Rank Updatescited by 200
  9. 2022Generating Training Data with Language Models: Towards Zero-Shot Language Understandingcited by 199
  10. 2021Adversary Instantiation: Lower Bounds for Differentially Private Machine Learningcited by 193
  11. 2021Data Augmentation Can Improve Robustnesscited by 193
  12. 2022Datamodels: Predicting Predictions from Training Datacited by 186
  13. 2022Dataset Distillation using Neural Feature Regressioncited by 177
  14. 2022Surrogate Gap Minimization Improves Sharpness-Aware Trainingcited by 171
  15. 2020Spectral Bias and Task-Model Alignment Explain Generalization in Kernel Regression and Infinitely Wide Neural Networkscited by 166
  16. 2021DICE: Leveraging Sparsification for Out-of-Distribution Detectioncited by 166
  17. 2021FSDR: Frequency Space Domain Randomization for Domain Generalizationcited by 163
  18. 2022Neural Networks can Learn Representations with Gradient Descentcited by 161
  19. 2021Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Pathcited by 156
  20. 2021The Low-Rank Simplicity Bias in Deep Networkscited by 150
  21. 2021Poisoning and Backdooring Contrastive Learningcited by 149
  22. 2021Dash: Semi-Supervised Learning with Dynamic Thresholdingcited by 147
  23. 2023Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothingcited by 147
  24. 2022Towards Understanding Sharpness-Aware Minimizationcited by 145
  25. 2024WARM: On the Benefits of Weight Averaged Reward Modelscited by 145
  26. 2022Learning Causally Invariant Representations for Out-of-Distribution Generalization on Graphscited by 144
  27. 2022Reconstructing Training Data with Informed Adversariescited by 143
  28. 2022Leveraging Unlabeled Data to Predict Out-of-Distribution Performancecited by 141
  29. 2022Training Data Influence Analysis and Estimation: A Surveycited by 134
  30. 2021Comparing Kullback-Leibler Divergence and Mean Squared Error Loss in Knowledge Distillationcited by 131
  31. 2021Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolationcited by 130
  32. 2021Co-Mixup: Saliency Guided Joint Mixup with Supermodular Diversitycited by 127
  33. 2021Learning curves of generic features maps for realistic datasets with a teacher-student modelcited by 126
  34. 2023On Memorization in Diffusion Modelscited by 122
  35. 2021Epistemic Neural Networkscited by 120
  36. 2023The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reductioncited by 120
  37. 2022REPAIR: REnormalizing Permuted Activations for Interpolation Repaircited by 119
  38. 2021Assessing Generalization of SGD via Disagreementcited by 116

50 most cited of 1,657 All 1,657 (JSON)