mlsanityData to 13 Oct 2026

A Geometric Framework for Understanding Memorization in Generative Models

arXiv 2411.00113 ↗First posted 31 Oct 2024stat.ML cs.LGcited by 40

Not on the map. Why?

Cited by 33 in Diffusion models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

As deep generative models have progressed, recent work has shown them to be capable of memorizing and reproducing training datapoints when deployed. These findings call into question the usability of generative models, especially in light of the legal and privacy risks brought about by memorization. To better understand this phenomenon, we propose the manifold memorization hypothesis (MMH), a geometric framework which leverages the manifold hypothesis into a clear language in which to reason about memorization. We propose to analyze memorization in terms of the relationship between the dimensionalities of (i) the ground truth data manifold and (ii) the manifold learned by the model. This framework provides a formal standard for "how memorized" a datapoint is and systematically categorizes memorized data into two types: memorization driven by overfitting and memorization driven by the underlying data distribution. By analyzing prior work in the context of the MMH, we explain and unify assorted observations in the literature. We empirically validate the MMH using synthetic data and image datasets up to the scale of Stable Diffusion, developing new tools for detecting and preventing generation of memorized samples in the process.

Key facts. Cited by 40 papers in the map (data through 13 Oct 2026). Builds on 50 of its 74 references that are in the map. First posted on arXiv 31 Oct 2024.

Builds on

50 of 74 references.

References · 50

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,588
  2. 2020Denoising Diffusion Probabilistic Modelscited by 20,952
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,777
  4. 2014Microsoft COCO: Common Objects in Contextcited by 17,550
  5. 2020Denoising Diffusion Implicit Modelscited by 9,718
  6. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,028
  7. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,023
  8. 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 6,000
  9. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,141
  10. 2021Improved Denoising Diffusion Probabilistic Modelscited by 3,751
  11. 2020Training Generative Adversarial Networks with Limited Datacited by 1,184
  12. 2014NICE: Non-linear Independent Components Estimationcited by 1,054
38 more
  1. 2022Diffusion-LM Improves Controllable Text Generationcited by 1,012
  2. 2022Quantifying Memorization Across Neural Language Modelscited by 763
  3. 2023Extracting Training Data from Diffusion Modelscited by 729
  4. 2021Score-based Generative Modeling in Latent Spacecited by 635
  5. 2022Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Modelscited by 421
  6. 2023Understanding and Mitigating Copying in Diffusion Modelscited by 208
  7. 2023Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion modelscited by 200
  8. 2019Diagnosing and Enhancing VAE Modelscited by 193
  9. 2023Mixed-Type Tabular Data Synthesis with Score-based Diffusion in Latent Spacecited by 184
  10. 2023Generalization in diffusion models arises from geometry-adaptive harmonic representationscited by 181
  11. 2022A Self-Supervised Descriptor for Image Copy Detectioncited by 176
  12. 2023On Memorization in Diffusion Modelscited by 123
  13. 2022Score-Based Generative Models Detect Manifoldscited by 122
  14. 2023On Provable Copyright Protection for Generative Modelscited by 97
  15. 2024Detecting, Explaining, and Mitigating Memorization in Diffusion Modelscited by 95
  16. 2023On the Generalization Properties of Diffusion Modelscited by 67
  17. 2022Verifying the Union of Manifolds Hypothesis for Image Datacited by 62
  18. 2023A Reproducible Extraction of Training Images from Diffusion Modelscited by 61
  19. 2020A Non-Parametric Test to Detect Data-Copying in Generative Modelscited by 52
  20. 2024Towards Memorization-Free Diffusion Modelscited by 50
  21. 2024Deep Generative Models through the Lens of the Manifold Hypothesis: A Survey and New Connectionscited by 47
  22. 2024Consistent Diffusion Meets Tweedie: Training Exact Ambient Diffusion Models with Noisy Datacited by 44
  23. 2024A Geometric View of Data Complexity: Efficient Local Intrinsic Dimension Estimation with Diffusion Modelscited by 41
  24. 2024Finding NeMo: Localizing Neurons Responsible For Memorization in Diffusion Modelscited by 37
  25. 2022Your diffusion model secretly knows the dimension of the data manifoldcited by 36
  26. 2024A Good Score Does not Lead to A Good Generative Modelcited by 36
  27. 2022When do GANs replicate? On the choice of dataset sizecited by 34
  28. 2022LIDL: Local Intrinsic Dimension Estimation Using Approximate Likelihoodcited by 30
  29. 2023On the Generalization of Diffusion Modelcited by 30
  30. 2022Diagnosing and Fixing Manifold Overfitting in Deep Generative Modelscited by 28
  31. 2021This Person (Probably) Exists. Identity Membership Attacks Against GAN Generated Facescited by 25
  32. 2023Beware of diffusion models for synthesizing medical images -- A comparison with GANs in terms of memorizing brain MRI and chest x-ray imagescited by 23
  33. 2024On gauge freedom, conservativity and intrinsic dimensionality estimation in diffusion modelscited by 22
  34. 2024A Geometric Explanation of the Likelihood OOD Detection Paradoxcited by 22
  35. 2023Mathematical analysis of singularities in the diffusion model under the submanifold assumptioncited by 20
  36. 2023Learning Manifold Dimensions with Conditional Variational Autoencoderscited by 20
  37. 2021On Training Sample Memorization: Lessons from Benchmarking Generative Modeling with a Large-scale Competitioncited by 18
  38. 2023Data-Copying in Generative Models: A Formal Frameworkcited by 9

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2024: 1; 2025: 21; 2026: 18.

40 papers.

Sort:

By area:

  1. 2025On the Closed-Form of Flow Matching: Generalization Does Not Arise from Target Stochasticitycited by 50
  2. 2025Memorization to Generalization: Emergence of Diffusion Models from Associative Memorycited by 35
  3. 2025On the Edge of Memorization in Diffusion Modelscited by 22
  4. 2024Understanding and Mitigating Memorization in Generative Models via Sharpness of Probability Landscapescited by 20
  5. 2025Denoising Score Matching with Random Features: Insights on Diffusion Models from Precise Learning Curvescited by 20
  6. 2025A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspectivecited by 18
  7. 2025Does Generation Require Memorization? Creative Diffusion Models using Ambient Diffusioncited by 17
  8. 2025A solvable model of learning generative diffusion: theory and insightscited by 14
  9. 2025TCFG: Tangential Damping Classifier-free Guidancecited by 14
  10. 2025Provable Separations between Memorization and Generalization in Diffusion Modelscited by 14
  11. 2025Carr'e du champ flow matching: better quality-generalisation tradeoff in generative modelscited by 13
  12. 2025Generalization of Diffusion Models Arises with a Balanced Representation Spacecited by 10
28 more
  1. 2025Resolving Memorization in Empirical Diffusion Model for Manifold Data in High-Dimensional Spacescited by 8
  2. 2025When Diffusion Models Memorize: Inductive Biases in Probability Flow of Minimum-Norm Shallow Neural Netscited by 6
  3. 2025How Diffusion Models Memorizecited by 6
  4. 2025Diffusion models under low-noise regimecited by 5
  5. 2026Local Coverage Governs Memorization in Diffusion Modelscited by 5
  6. 2025Generative AI Training and Copyright Lawcited by 3
  7. 2025Bob's Confetti: Phonetic Memorization Attacks in Music and Video Generationcited by 3
  8. 2026Biased Generalization in Diffusion Modelscited by 3
  9. 2026Asymptotic Learning Curves for Diffusion Models with Random Features Score and Manifold Datacited by 3
  10. 2025On Convolutions, Intrinsic Dimension, and Diffusion Modelscited by 2
  11. 2025Review of Hallucination Understanding in Large Language and Vision Modelscited by 2
  12. 2026Detecting and Mitigating Memorization in Diffusion Models through Anisotropy of the Log-Probabilitycited by 2
  13. 2026Rethinking Input Domains in Physics-Informed Neural Networks via Geometric Compactification Mappingscited by 2
  14. 2026Manifold Generalization Provably Proceeds Memorization in Diffusion Modelscited by 2
  15. 2025Dominant vs. Dominated: Concept-Level Generative Collapse in Diffusion Modelscited by 1
  16. 2026Reducing Memorisation in Generative Models via Riemannian Bayesian Inferencecited by 1
  17. 2026Local Hessian Spectral Filtering for Robust Intrinsic Dimension Estimationcited by 1
  18. 2026Local Intrinsic Dimension Unveils Hallucinations in Diffusion Modelscited by 1
  19. 2026Conf-Gen: Conformal Uncertainty Quantification for Generative Modelscited by 1
  20. 2025Exploring Image Generation via Mutually Exclusive Probability Spaces and Local Correlation Hypothesiscited by 0
  21. 2026Memorization In Stable Diffusion Is Unexpectedly Driven by CLIP Embeddingscited by 0
  22. 2026Filtering Memorization from Parameter-Space in Diffusion Modelscited by 0
  23. 2026Localizing Memorized Regions in Diffusion Models via Coordinate-Wise Curvature Differencescited by 0
  24. 2026Fingerprinting Text-to-Image Diffusion Models via Collapsed Generationcited by 0
  25. 2026Background Gradients Shape Memorization in Flow Matchingcited by 0
  26. 2026From Modes to Memories: Characterizing the Scale-Space Dynamics of Diffusion Modelscited by 0
  27. 2026The Null Is the Hard Part: Exact Tests for Memorization in Generative Modelscited by 0
  28. 2026Rethinking Memorization Mitigation in Diffusion Models: Reinforcing Text Conditioningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2024Losing dimensions: Geometric memorization in generative diffusion
  2. 2025Memorization in 3D Shape Generation: An Empirical Study
  3. 2025Unconsciously Forget: Mitigating Memorization; Without Knowing What is being Memorized
  4. 2024What Secrets Do Your Manifolds Hold? Understanding the Local Geometry of Generative Models
  5. 2023Understanding (Un)Intended Memorization in Text-to-Image Generative Models
  6. 2026Two Calm Ends and the Wild Middle: A Geometric Picture of Memorization in Diffusion Models
  7. 2026Diffusion Models Preferentially Memorize Prototypical Examples or: Why Does My Diffusion Model Love Slop?
  8. 2024An Inversion-based Measure of Memorization for Diffusion Models