Bigger Isn't Always Memorizing: Early Stopping Overparameterized Diffusion Models
cs.LG, stat.ML
Submitted: 2025-05-22
Updated: 2026-11-18
Code: https://github.com/openai/improved-diffusion
Terminology
Sources
- Hierarchical Text-Conditional Image Generation with CLIP Latents
- Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets
- Simple and Effective Masked Diffusion Language Models
- Simplified and Generalized Masked Diffusion for Discrete Data
- Large Language Diffusion Models
- Score-Based Generative Modeling through Stochastic Differential Equations
- Membership Inference Attacks Against Text-to-image Generation Models
- Membership Inference of Diffusion Models
- The Emergence of Reproducibility and Generalizability in Diffusion Models
- Generalization in diffusion models arises from geometry-adaptive harmonic representations
- How Compositional Generalization and Creativity Improve as Diffusion Models are Trained
- Differentially Private Diffusion Models
- PixelCNN++: Improving the PixelCNN with Discretized Logistic Mixture Likelihood and Other Modifications
- Neumann Optimizer: A Practical Optimization Algorithm for Deep Neural Networks
- Why Diffusion Models Don't Memorize: The Role of Implicit Dynamical Regularization in Training
- A Phase Transition in Diffusion Models Reveals the Hierarchical Nature of Data
- Probing the Latent Hierarchical Structure of Data via Diffusion Models
- Feature Learning in Infinite-Width Neural Networks
- Learning curves theory for hierarchically compositional data with power-law distributed features
- The Unreasonable Effectiveness of Gaussian Score Approximation for Diffusion Models and its Applications
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis