Prof. Ricard Gavaldà (Universitat Politècnica de Catalunya) is visiting us this week. He will give a talk on Thursday (November 9th), at 11:30, in room Henri Poincaré on Tensor decomposition for inferring latent variable models, and an application to healthcare analytics.
Abstract: We present a new tensor decomposition method for inferring the parameters of certain kinds of latent variable models. The method has a number of advantages over existing methods, including scalability, the fact that it is deterministic and not randomized, and the quality of the results. It is also particularly appropriate for clustering tasks. In this respect, we present an application to clustering patients on the basis of their sets of diagnostics, and obtain clustering on two datasets provided by the Catalan Service of Health that clinicians find meaningful and potentially useful. This is joint work with Matteo Ruffini and Esther Limón.