[ABE-L] Seminários de Probabilidade e Sistemas Complexos, ICMC-USP & UFSCar. 12/06 (16h00)

[Pablo Martin Rodriguez]

Prezados colegas e alunos, 

Esta semana a palestra do ciclo de Seminários de Probabilidade e Sistemas Complexos , ICMC-USP & UFSCar, São Carlos, será realizada na sexta 12/06 na UFSCar. Uma lista das palestras passadas, e as próximas, pode ser encontrada no seguinte link: 

http://www.icmc.usp.br/pessoas/pablor/seminar/pcs_seminar.html 

Segue a informação da palestra . 
Nonparametric conditional density estimation in high dimensions: some theoretical aspects 
Speaker: Rafael Izbicki (UFSCar) 
Date: 12/06/2015 • Time: 16h00 • Room: Sala de Seminários DEs-UFSCar 

Abstract: In some applications (e.g., in cosmology and economics), the regression E[Z|x] is not adequate to represent the association between a predictor x and a response Z because of multi-modality and asymmetry of f(z|x); using the full density instead of a single-point estimate can then lead to less bias in subsequent analysis. However, there are currently no effective ways of estimating f(z|x) when x represents high-dimensional, complex data. Here we propose a fully nonparametric approach to conditional density estimation (CDE) that reformulates CDE as a non-parametric orthogonal series problem where the expansion coefficients are estimated by regression. By taking such an approach, one can efficiently estimate conditional densities in high dimensions by drawing upon the success in high-dimensional regression. Depending on the choice of regression procedure, our method can adapt to a variety of challenging high-dimensional settings with, for example, a large number of irrelevant components or nonlinear manifold structure in the data. We study the theoretical and empirical performance of our proposed method, and we compare our approach with traditional conditional density estimators on real-world as well as simulated data. 

Estão todos convidados! Favor divulgar a possíveis interessados. 

Um abraço, 
Pablo 
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <https://lists.ime.usp.br/archives/abe/attachments/20150609/bb01adf8/attachment.html>