Date/heure
Date(s) - 8 novembre 2013

Catégories Pas de Catégories

$Functional Data Analysis via Bayesian nonparametrics with application to signal classification By A. Rabaoui, Institut Fresnel, Marseille. In many signal processing applications, data collected are drawn from continuous processes and often obtained in the form of functions. Functional data analysis (FDA) is a very attractive field of research that provides the possibility to fully exploit structure in such inherently continuous data. While more traditional approaches of functional data analysis are parametric and require specifying in advance a basis function for the data, challenges in modern signal processing applications motivate the nonparametric analysis of these data. In this talk, I will show how to deal with functional classification problems from a Bayesian nonparametric approach. Then, I will provide theoretical and practical motivations for our approach using Dirichlet process mixtures and Gaussian processes. Finally, I will illustrate experimentally that the Bayesian nonparametric FDA framework is particularly relevant for signal processing applications where attributes (features) are really functions and can be dependent of each others.$