mlsanityData to 7 Oct 2026

Scaling Non-Parametric Sampling with Representation

arXiv 2510.22196 ↗First posted 25 Oct 2025cs.CV cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Scaling and architectural advances have produced strikingly photorealistic image generative models, yet their mechanisms still remain opaque. Rather than advancing scaling, our goal is to strip away complicated engineering tricks and propose a simple, non-parametric generative model. Our design is grounded in three principles of natural images-(i) spatial non-stationarity, (ii) low-level regularities, and (iii) high-level semantics-and defines each pixel's distribution from its local context window. Despite its minimal architecture and no training, the model produces high-fidelity samples on MNIST and visually compelling CIFAR-10 images. This combination of simplicity and strong empirical performance points toward a minimal theory of natural-image structure. The model's white-box nature also allows us to have a mechanistic understanding of how the model generalizes and generates diverse images. We study it by tracing each generated pixel back to its source images. These analyses reveal a simple, compositional procedure for "part-whole generalization", suggesting a hypothesis for how large neural network generative models learn to generalize.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 39 of its 58 references that are in the map. First posted on arXiv 25 Oct 2025.

Builds on

39 of 58 references.

References · 39

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  2. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  3. 2022Generative Adversarial Networkscited by 5,603
  4. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,067
  5. 2013Auto-Encoding Variational Bayescited by 3,980
  6. 2019Analyzing and Improving the Image Quality of StyleGANcited by 3,580
  7. 2022Elucidating the Design Space of Diffusion-Based Generative Modelscited by 2,994
  8. 2016Improved Techniques for Training GANscited by 2,938
  9. 2020Exploring Simple Siamese Representation Learningcited by 2,853
  10. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,048
  11. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,507
  12. 2022In-context Learning and Induction Headscited by 930
27 more
  1. 2022Toy Models of Superpositioncited by 928
  2. 2016Conditional Image Generation with PixelCNN Decoderscited by 859
  3. 2023Progress measures for grokking via mechanistic interpretabilitycited by 822
  4. 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 703
  5. 2019A Theoretical Analysis of Contrastive Unsupervised Representation Learningcited by 530
  6. 2021Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformerscited by 316
  7. 2021Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Losscited by 311
  8. 2019Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNetcited by 260
  9. 2021Understanding self-supervised Learning Dynamics without Contrastive Pairscited by 254
  10. 2022Contrastive and Non-Contrastive Self-Supervised Learning Recover Global and Local Spectral Embedding Methodscited by 146
  11. 2021Decoupled Contrastive Learningcited by 141
  12. 2022On the duality between contrastive and non-contrastive self-supervised learningcited by 107
  13. 2021Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factorscited by 106
  14. 2024An analytic theory of creativity in convolutional diffusion modelscited by 103
  15. 2021ReduNet: A White-box Deep Network from the Principle of Maximizing Rate Reductioncited by 84
  16. 2014PCANet: A Simple Deep Learning Baseline for Image Classification?cited by 70
  17. 2021High Fidelity Visualization of What Your Self-Supervised Representation Knows Aboutcited by 64
  18. 2018Banach Wasserstein GANcited by 56
  19. 2022On the Principles of Parsimony and Self-Consistency for the Emergence of Intelligencecited by 50
  20. 2022Neural Manifold Clustering and Embeddingcited by 44
  21. 2018The Sparse Manifold Transformcited by 36
  22. 2019Learning Energy-Based Models in High-Dimensional Spaces with Multi-scale Denoising Score Matchingcited by 26
  23. 2022Masked Siamese ConvNetscited by 22
  24. 2022What Do We Maximize in Self-Supervised Learning?cited by 15
  25. 2022Bag of Image Patch Embedding Behind the Success of Self-Supervised Learningcited by 9
  26. 2021The Unreasonable Effectiveness of Patches in Deep Convolutional Kernels Methodscited by 8
  27. 2022Revisiting the Critical Factors of Augmentation-Invariant Representation Learningcited by 6

Built on by

Not cited yet.