Matthieu Wyart

Bigger Isn’t Always Memorizing: Early Stopping Overparameterized Diffusion Models

On the Emergence of Linear Analogies in Word Embeddings

How Compositional Generalization and Creativity Improve as Diffusion Models are Trained

Probing the Latent Hierarchical Structure of Data via Diffusion Models

Learning curves theory for hierarchically compositional data with power-law distributed features