Differentiable Clustering with Perturbed Spanning Forests - Archive ouverte HAL Accéder directement au contenu
Pré-Publication, Document De Travail Année : 2023

Differentiable Clustering with Perturbed Spanning Forests

Lawrence Stewart
  • Fonction : Auteur
  • PersonId : 1184478
Francis S Bach
  • Fonction : Auteur
  • PersonId : 863086
Quentin Berthet
  • Fonction : Auteur

Résumé

We introduce a differentiable clustering method based on minimum-weight spanning forests, a variant of spanning trees with several connected components. Our method relies on stochastic perturbations of solutions of linear programs, for smoothing and efficient gradient computations. This allows us to include clustering in end-to-end trainable pipelines. We show that our method performs well even in difficult settings, such as datasets with high noise and challenging geometries. We also formulate an ad hoc loss to efficiently learn from partial clustering data using this operation. We demonstrate its performance on several real world datasets for supervised and semi-supervised tasks.
Fichier principal
Vignette du fichier
main.pdf (1.35 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04105310 , version 1 (24-05-2023)
hal-04105310 , version 2 (30-05-2023)
hal-04105310 , version 3 (05-11-2023)

Identifiants

  • HAL Id : hal-04105310 , version 1

Citer

Lawrence Stewart, Francis S Bach, Felipe Llinares López, Quentin Berthet. Differentiable Clustering with Perturbed Spanning Forests. 2023. ⟨hal-04105310v1⟩
114 Consultations
47 Téléchargements

Partager

Gmail Facebook X LinkedIn More