Pré-Publication, Document De Travail Année : 2016

A Quasi-Bayesian Perspective to Online Clustering

Résumé

When faced with high frequency streams of data, clustering raises theoretical and algorith-mic pitfalls. We introduce a new and adaptive online clustering algorithm relying on a quasi-Bayesian approach, with a dynamic (i.e., time-dependent) estimation of the (unknown and changing) number of clusters. We prove that our approach is supported by minimax regret bounds. We also provide an RJMCMC-flavored implementation (called PACBO) for which we give a convergence guarantee. Finally, numerical experiments illustrate the potential of our procedure.

Fichier principal
Vignette du fichier
main.pdf (548.45 Ko) Télécharger le fichier

Dates et versions

hal-01264233 , version 1 (28-01-2016)
hal-01264233 , version 2 (07-04-2017)
hal-01264233 , version 3 (08-04-2017)
hal-01264233 , version 4 (25-05-2018)

Licence

Identifiants

  • HAL Id : hal-01264233 , version 2

Citer

Le Li, Benjamin Guedj, Sébastien Loustau. A Quasi-Bayesian Perspective to Online Clustering. 2016. ⟨hal-01264233v2⟩
2224 Consultations
705 Téléchargements

Partager

  • More