Estimation of univariate Gaussian mixtures for huge raw datasets by using binned datasets

Filippo Antonazzo (1) , Christophe Biernacki (1) , Christine Keribin (2)

1 MODAL - MOdel for Data Analysis and Learning
2 CELESTE - Statistique mathématique et apprentissage

Popularity of unsupervised learning is magnified by the regular increase of sample sizes. Indeed, it provides opportunity to reveal information previously out of scope. However, the volume of data leads to some issues related to prohibitive calculation times and also to high energy consumption and the need of high computational ressources. Resorting to binned data depending on an adaptive grid is expected to give proper answer to such green computing issues while not harming the related estimation issues. A first attempt is conducted in the context of univariate Gaussian mixtures, included a numerical illustration and some theoretical advances.

Mots clés: Big data; Binned data; Unsupervised learning; Green computing

Domaines: Statistiques [stat]

Filippo Antonazzo, Christophe Biernacki, Christine Keribin. Estimation of univariate Gaussian mixtures for huge raw datasets by using binned datasets. JDS 2020 - 52ème Journées de Statistiques de la Société Française de Statistique, May 2020, Nice, France. ⟨hal-03082437v2⟩