Data to 7 Oct 2026

Does Learning Require Memorization? A Short Tale about a Long Tail

arXiv 1906.05271 ↗First posted 12 Jun 2019cs.LG stat.MLcited by 452

Not on the map. Why?

Cited by 74 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

State-of-the-art results on image recognition tasks are achieved using over-parameterized learning algorithms that (nearly) perfectly fit the training set and are known to fit well even random labels. This tendency to memorize the labels of the training data is not explained by existing theoretical analyses. Memorization of the training data also presents significant privacy risks when the training data contains sensitive personal information and thus it is important to understand whether such memorization is necessary for accurate learning. We provide the first conceptual explanation and a theoretical model for this phenomenon. Specifically, we demonstrate that for natural data distributions memorization of labels is necessary for achieving close-to-optimal generalization error. Crucially, even labels of outliers and noisy labels need to be memorized. The model is motivated and supported by the results of several recent empirical works. In our model, data is sampled from a mixture of subpopulations and our results show that memorization is necessary whenever the distribution of subpopulation frequencies is long-tailed. Image and text data is known to be long-tailed and therefore our results establish a formal link between these empirical phenomena. Our results allow to quantify the cost of limiting memorization in learning and explain the disparate effects that privacy and model compression have on different subgroups.

Key facts. Cited by 452 papers in the map (data through 7 Oct 2026). First posted on arXiv 12 Jun 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 50; 2022: 78; 2023: 85; 2024: 90; 2025: 82; 2026: 67.

452 papers.

Sort:

By area:

2023PaLM 2 Technical Reportcited by 1,154
2021Membership Inference Attacks From First Principlescited by 839
2023Extracting Training Data from Diffusion Modelscited by 723
2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 529
2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 418
2021Differentially Private Fine-tuning of Language Modelscited by 392
2023Detecting Pretraining Data from Large Language Modelscited by 375
2023Analyzing Leakage of Personally Identifiable Information in Language Modelscited by 278
2024RAFT: Adapting Language Model to Domain Specific RAGcited by 258
2021Enhanced Membership Inference Attacks against Machine Learning Modelscited by 237
2022Memorization Without Overfitting: Analyzing the Training Dynamics of Large Language Modelscited by 218
2022What Does it Mean for a Language Model to Preserve Privacy?cited by 194

38 more

2022Can Bad Teaching Induce Forgetting? Unlearning in Deep Networks using an Incompetent Teachercited by 189
2022Datamodels: Predicting Predictions from Training Datacited by 186
2023Fast Machine Unlearning Without Retraining Through Selective Synaptic Dampeningcited by 177
2021On the Necessity of Auditable Algorithmic Definitions for Machine Unlearningcited by 164
2021The Curious Case of Hallucinations in Neural Machine Translationcited by 161
2021Counterfactual Memorization in Neural Language Modelscited by 155
2022Reconstructing Training Data with Informed Adversariescited by 143
2022Impact of Pretraining Term Frequencies on Few-Shot Reasoningcited by 142
2022Training Data Influence Analysis and Estimation: A Surveycited by 134
2021Scaling Up Influence Functionscited by 130
2022Reconstructing Training Data from Trained Neural Networkscited by 129
2022The Privacy Onion Effect: Memorization is Relativecited by 128
2022Zero-Shot Machine Unlearningcited by 125
2023On Memorization in Diffusion Modelscited by 122
2021Large-Scale Differentially Private BERTcited by 117
2021Interpretable Deep Learning: Interpretation, Interpretability, Trustworthiness, and Beyondcited by 116
2021Dataset Inference: Ownership Resolution in Machine Learningcited by 115
2023Calibrated Language Models Must Hallucinatecited by 114
2022Neural Collapse: A Review on Modelling Principles and Generalizationcited by 95
2022Truth Serum: Poisoning Machine Learning Models to Reveal Their Secretscited by 94
2021Trustworthy AI: A Computational Perspectivecited by 93
2024Investigating Data Contamination for Pre-training Language Modelscited by 91
2023Practical Membership Inference Attacks against Fine-tuned Large Language Models via Self-prompt Calibrationcited by 90
2024What makes unlearning hard and what to do about itcited by 89
2025MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbationscited by 88
2022Position: Considerations for Differentially Private Learning with Large-Scale Public Pretrainingcited by 86
2024How Do Large Language Models Acquire Factual Knowledge During Pretraining?cited by 86
2024Generalization v.s. Memorization: Tracing Language Models' Capabilities Back to Pretraining Datacited by 83
2022Distilling Model Failures as Directions in Latent Spacecited by 82
2024Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Modelscited by 78
2024Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systemscited by 76
2024DE-COP: Detecting Copyrighted Content in Language Models Training Datacited by 72
2023Bag of Tricks for Training Data Extraction from Language Modelscited by 67
2023Can Neural Network Memorization Be Localized?cited by 66
2021Clusterability as an Alternative to Anchor Points When Learning with Noisy Labelscited by 65
2024When Do We Not Need Larger Vision Models?cited by 65
2022Understanding Transformer Memorization Recall Through Idiomscited by 62
2021On Memorization in Probabilistic Deep Generative Modelscited by 60

50 most cited of 452 All 452 (JSON)