Skip to Main content Skip to Navigation

Transport Optimal pour l'Apprentissage Automatique

Abstract : This thesis proposes theoretical and numerical contributions to use Entropy-regularized Optimal Transport (EOT) for machine learning. We introduce Sinkhorn Divergences (SD), a class of discrepancies between probability measures based on EOT which interpolates between two other well-known discrepancies: Optimal Transport (OT) and Maximum Mean Discrepancies (MMD). We develop an ecient numerical method to use SD for density fitting tasks, showing that a suitable choice of regularization can improve performance over existing methods. We derive a sample complexity theorem for SD which proves that choosing a large enough regularization parameter allows to break the curse of dimensionality from OT, and recover asymptotic rates similar to MMD.We propose and analyze stochastic optimization solvers for EOT, which yield online methods that can cope with arbitrary measures and are well suited to large scale problems, contrarily to existing discrete batch solvers.
Document type :
Complete list of metadatas

Cited literature [90 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Tuesday, January 28, 2020 - 2:23:11 PM
Last modification on : Wednesday, September 23, 2020 - 4:40:44 AM
Long-term archiving on: : Wednesday, April 29, 2020 - 3:14:08 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02458044, version 1



Aude Genevay. Transport Optimal pour l'Apprentissage Automatique. Machine Learning [stat.ML]. Université Paris sciences et lettres, 2019. Français. ⟨NNT : 2019PSLED002⟩. ⟨tel-02458044⟩



Record views


Files downloads