The balance between feature learning and collapse in dynamical systems-based generative models
stat.ML, cs.AI, cs.LG, math.DS
Submitted: 2026-11-09
Updated: 2026-11-09
Terminology
Sources
- Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets
- Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
- A theoretical perspective on mode collapse in variational inference
- Dynamics of stochastic gradient descent for two-layer neural networks in the teacher-student setup
- Score-Based Generative Modeling through Stochastic Differential Equations
- Proteina: Scaling Flow-based Protein Structure Generative Models
- Beyond ELBOs: A Large-Scale Evaluation of Variational Methods for Sampling
- Class-Balancing Diffusion Models
- Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
- KPFlow: An Operator Perspective on Dynamic Collapse Under Gradient Descent Training of Recurrent Networks
- Deep learning generalizes because the parameter-function map is biased towards simple functions
- Saddle-To-Saddle Dynamics in Deep ReLU Networks: Low-Rank Bias in the First Saddle Escape
- Generalization Dynamics of Linear Diffusion Models
- A theory of learning data statistics in diffusion models, from easy to hard
- One-Step Offline Distillation of Diffusion-based Models via Koopman Modeling
- Hierarchical Koopman Diffusion: Fast Generation with Interpretable Diffusion Trajectory
- Avoiding mode collapse in diffusion models fine-tuned with reinforcement learning
- Training Neural Samplers with Reverse Diffusive KL Divergence
- Heavy-tailed denoising score matching
- Generalization in diffusion models arises from geometry-adaptive harmonic representations
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