Generative diffusion in very large dimensions

Abstract

Generative models based on diffusion have become the state of the art in the last few years, notably for image generation. Here, we analyse them in the high-dimensional limit, where data are formed by a very large number of variables. We use methods from statistical physics and focus on two well-controlled high-dimensional cases: a Gaussian model and the Curie-Weiss model of ferromagnetism. In the latter case, we highlight the mechanism of symmetry breaking in the inverse diffusion, and point out that, in order to reconstruct the relative asymmetry of the two low-temperature states, and thus to obtain the correct probability weights, one needs a database with a number of points much larger than the dimension of each data point. We characterize the scaling laws in the number of data and in the number of dimensions for an efficient generation.

arXiv comment10 pages

References in corpus (11)

• Denoising Diffusion Probabilistic Models
• Score-Based Generative Modeling through Stochastic Differential Equations
• Generative Modeling by Estimating Gradients of the Data Distribution
• Statistical Aspects of Wasserstein Distances
• From non-ergodic eigenvectors to local resolvent statistics and back: a random matrix perspective
• Non-Ergodic Delocalization in the Rosenzweig-Porter Model
• Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling
• Eigenvectors distribution and quantum unique ergodicity for deformed Wigner matrices
• Convergence of denoising diffusion models under the manifold hypothesis
• Wavelet Score-Based Generative Modeling
• Convergence for score-based generative modeling with polynomial complexity

Cited by in corpus (15)

• Dynamical Regimes of Diffusion Models
• Sampling with flows, diffusion and autoregressive neural networks: A spin-glass perspective
• Spin glass theory and its new challenge: structured disorder
• Nonreciprocal Spin-Glass Transition and Aging
• Nonequilbrium physics of generative diffusion models
• Performance of machine-learning-assisted Monte Carlo in sampling from simple statistical physics models
• A Very Effective and Simple Diffusion Reconstruction for the Diluted Ising Model
• Analysis of Diffusion Models for Manifold Data
• U-Turn Diffusion
• The Wishart--Rosenzweig--Porter random matrix ensemble
• Geometric Regularity in Deterministic Sampling Dynamics of Diffusion-based Generative Models
• Markov spin models for image generation : explicit large deviations with respect to the number of pixels
• Convergence properties of Markov models for image generation with applications to spin-flip dynamics and to diffusion processes
• The Information Dynamics of Generative Diffusion
• Dynamical Regimes of Discrete Diffusion Models