Clustering and classification through normalizing flows in feature space.

JUAN PABLO AGNELLI; ESTEBAN TABAK; MARTIN CADEIRAS; CRISTINA TURNER; ERIC VANDEN-EIJNDEN

MULTISCALE MODELING AND SIMULATION

SIAM PUBLICATIONS

Lugar: Philadelphia-USA; Año: 2010 vol. 8 p. 1784 - 1802

1540-3459

A unified variational methodology is developed for classification and clustering problems, and tested in classification of tumors from gene expression data. It is based on fluid-like flows in feature space that cluster a set of observations by transforming them into likely samples from p isotropics Gaussians, where p is the number of classes sought. The methodology blurs the distinction between training a testing populations through the soft assignment of both to classes. The observations acts as Lagrangian markers for the flows, comparatively active or passive depending on the current strenght of the assignment to the corresponding class.