Breakdown of Local Denoising as Semantic Speciation
cs.LG, cond-mat.dis-nn, cond-mat.stat-mech
Submitted: 2026-09-29
Updated: 2026-09-29
Comments: 9 pages main, 13 pages appendix, 4 figures. Comments very welcome
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- How Out-of-Equilibrium Phase Transitions can Seed Pattern Formation in Trained Diffusion Models
- Diffusion Models Beat GANs on Image Synthesis
- A General Derivative Identity for the Conditional Mean Estimator in Gaussian Noise and Some Applications
- Scaling Rectified Flow Transformers for High-Resolution Image Synthesis
- Measuring Semantic Information Production in Generative Diffusion Models
- The Entropic Signature of Class Speciation in Diffusion Models
- Classifier-Free Diffusion Guidance
- Local Diffusion Models and Phases of Data Distributions
- An exact information theory of generalization phase transitions in Bayesian diffusion models
- An analytic theory of creativity in convolutional diffusion models
- Locality in Image Diffusion Models Emerges from Data Statistics
- Towards a Mechanistic Explanation of Diffusion Model Generalization
- On Variational Bounds of Mutual Information
- Does Generation Require Memorization? Creative Diffusion Models using Ambient Diffusion
- Dynamical Regimes of Discrete Diffusion Models
- Convexity of mutual information along the heat flow
- MMG: Mutual Information Estimation via the MMSE Gap in Diffusion
- Conditional Mutual Information and Information-Theoretic Phases of Decohered Gibbs States
- Concurrence of Symmetry Breaking and Nonlocality Phase Transitions in Diffusion Models
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks