The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models
stat.ML, cond-mat.dis-nn, cs.LG
Submitted: 2026-05-07
Updated: 2026-11-11
Terminology
Sources
- Losing dimensions: Geometric memorization in generative diffusion
- The statistical thermodynamics of generative diffusion models: Phase transitions, symmetry breaking and critical instability
- A solvable model of learning generative diffusion: theory and insights
- Analysis of learning a flow-based generative model from limited sample complexity
- Bigger Isn't Always Memorizing: Early Stopping Overparameterized Diffusion Models
- Denoising Score Matching with Random Features: Insights on Diffusion Models from Precise Learning Curves
- On Memorization in Diffusion Models
- Stage-wise Dynamics of Classifier-Free Guidance in Diffusion Models
- Adam: A Method for Stochastic Optimization
- Diffusion Models already have a Semantic Latent Space
- SGDR: Stochastic Gradient Descent with Warm Restarts
- Decoupled Weight Decay Regularization
- A solvable high-dimensional model where nonlinear autoencoders learn structure invisible to PCA while test loss misaligns with generalization
- Generalization Dynamics of Linear Diffusion Models
- Dynamical Decoupling of Generalization and Overfitting in Large Two-Layer Networks
- Breaking the Curse of Dimensionality: Diffusion Models Efficiently Learn Low-Dimensional Distributions
- Non-Uniform Diffusion Models
- Exploring Bias in over 100 Text-to-Image Generative Models
- An Analytical Theory of Spectral Bias in the Learning Dynamics of Diffusion Models
- Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
Related papers
- Uniform-in-Time Weak Propagation-of-Chaos in Shallow Neural Networks
- Diffusion Models Observe Only Gradients: A Geometric Perspective on Score Matching Errors
- Quantifying Error Propagation and Model Collapse in Diffusion Models
- Behavior of prediction performance metrics with rare events
- Statistical Inference and Learning for Shapley Additive Explanations (SHAP)
- Control Variate Design for Score-Function Gradient Estimators in Importance-Weighted Variational Inference