Note: Some of the graphical elements of this site are only visible to browsers that support accepted web standards. The content of this site is, however, accessible to any browser or Internet device.
Seminar Details
Speaker:
Dr. Shiro Ikeda,
Institute of Statistical Mathematics.
Venue:
Gabor Seminar Room
Time:
13 November 2003, Thursday, 1.00 PM
Title:
Two Extensions of Independent Component Analysis
Abstract:
ICA (Independent Component Analysis) is a powerful tool to analyze multivariate
data. In this talk, I introduce two extensions of ICA methods. The first
approach focuses on separatation of noise-contaminated data for speech
separation applications. For noisy data, we have combined factor analysis
with ICA. In the process of factor analysis, the number of the sources
and the amount of the noises are estimated. After factor analysis, an
ICA based method is further applied. We show that this approach is effective
for MEG data.
For speech separation, we have developed an ICA method in time-frequency
domain. Time-frequency ICA brings two new problems, that is, permutation
and amplitude ambiguities. We have developed a method to solve those problems,
and successfully separated speech signals recorded in real environment.
Bio:
Shiro Ikeda obtained his first degree from the University of Tokyo in
1991, and received his Ph.D in 1996 from the Department of Mathematical
Engineering and Information Physics, University of Tokyo.
From 1996, he worked as a postdoctoral reseracher at RIKEN, Tokyo, with
S. Amari, and in 2001 moved to Kyushu Institute of Technology as an Associate
Professor. In 2003, he joined the Institute of Statistical Mathematics,
Tokyo, which is his current affiliation. At the moment he is at UCL, London,
under the fellowship of the Royal Society.
Contact Info / Links:
Dr Shiro Ikeda's website: http://www.ism.ac.jp/~shiro/