marco cuturi

Completed Hyperparameter Transfer across Modules, Width, Depth, Batch and Duration

Flow Matching with Semidiscrete Couplings

The data-quality illusion: Rethinking Classifier-based Quality Filtering for LLM Pretraining

Addressing Misspecification in Simulation-based Inference through Data-driven Calibration

Controlling Language and Diffusion Models by Transporting Activations

Sample and Map from a Single Convex Potential: Generation using Conjugate Moment Measures

On Fitting Flow Models with Large Sinkhorn Couplings

Simple ReFlow: Improved Techniques for Fast Flow Models

LinEAS: End-to-end Learning of Activation Steering with a Distributional Loss

Multivariate Conformal Prediction using Optimal Transport

Scaling Laws for Forgetting during Finetuning with Pretraining Data Injection

Addressing Misspecification in Simulation-based Inference through Data-driven Calibration

Sparse Repellency for Shielded Generation in Text-to-Image Diffusion Models

Disentangled Representation Learning with the Gromov-Monge Gap

Shielded Diffusion: Generating Novel and Diverse Images using Sparse Repellency