Kernel based discriminant analysis and noisy ICA models for

advertisement
KATHOLIEKE UNIVERSITEIT LEUVEN
Statistics Seminar
Joint organization Statistics, Operations Research & Business Statistics,
Econometrics and Actuarial Sciences research groups
University Center for Statistics
Professor Dr Anestis Antoniadis
Université Joseph Fourier
Grenoble, France
“Kernel based discriminant analysis and noisy ICA models for
dimensionality reduction”
Thursday March 9, 2006
12:00—13:00
Room: PW.00.0001, Chemische Ingenieurstechnieken, W. de Croylaan 46, Heverlee
Abstract:
We propose a nonparametric discrimination method based on a nonparametric kernel
type-estimator of the posterior probability that an incoming observed vector is in a given
class. To overcome the curse of dimensionality of the multivariate kernel density
estimate, we assume that the data have been generated by a noisy independent component
model. By assuming such a model, the multivariate kernel estimators are replaced by
univariate kernel product estimators and the approach therefore provides a dimension
reduction together with semiparametric estimates of the class conditional probability
density functions. This density approximation is plugged into the classic Bayes rule and
its performance is evaluated both on real and simulated data.
Joint work (in progress) with U. Amato, A. Antoniadis, A. Samarov and A. Tsybakov
Prof. Antoniadis will also give a talk on “Low-level processing of mass spectrometry data and wavelet-based
mixed effects models for their analysis”, on March 21, from 12:15 till 13:15, at the Psychologisch Instituut,
KULeuven, Room C 00.60
Download