Independent Component Analysis Algorithms and Applications

Dr. Te-Won Lee

Salk Institute, La Jolla, CA

July 29, 1999 2:00 pm

Independent Component Analysis (ICA) is a method for extracting independent sour
ces given only mixtures of the unknown sources. This method has a wide range of 
applications in signal- and image processing issues and it has been suggested th
at ICA may reveal information about sensory coding strategies in the brain.

The goal of my talk is to introduce ICA algorithms, to give examples of applicat
ions in biomedical data analysis, and to summarize my most recent work in ICA mi
xture models.

I will present the assumptions in ICA and explain how algorithms can be derived 
from different theoretical principles such as information maximization, maximum 
likelihood and cummulants maximization. The algorithms are applied to sensory in
formation from the brain such as electroencephalographic data and functional mag
netic resonance imaging data. 
The results suggest that ICA can extract significant signals that were hidden in
the observed data. The ICA mixture model is a way of extending the ICA formulati
on to classification problems. It can be applied to learn the underlying structu
res of images and sounds. The obtained image codes are used for image processing
issues such as unsupervised classification, segmentation, de-noising, compressio
n, and hyperspectral color image representation. The ICA codes learned for natur
al images resemble the localized and oriented receptive fields found in the visu
al cortex of mammals and the codes for hyperspectral images are related to human
cone sensitivities. For speech signals we obtain codes that can be used for robu
st speech recognition.

ICA provides an interesting approach to data representation problems as well as 
new insights for coding sensory information in the brain.

Te-Won Lee

Te-Won Lee was born in Chungnam, Korea in 1969.  He received his diploma degree 
in March 1995 and his Ph.D. degree in October 1997 with highest honor in electri
cal engineering from the University of Technology Berlin. He was a visiting grad
uate student at the Institute Nationale Polytechnique de Grenoble, the Universit
y of California at Berkeley, and the Carnegie Mellon University.  
From 1995 to 1997 he was a Max-Planck Institute fellow.  He is currently a Resea
rch Associate with the Computational Neurobiology Laboratory at The Salk Institu
te and a Research Assistant Professor at the Institute for Neural Computation at
 the University of California, San Diego. 
His research interests are in the areas of unsupervised learning algorithms, art
ificial neural networks and Bayesian probability theory with applications to bio
medical signal processing and data mining.

This page is maintained by InYoung Kim (
Last update: October 2, 2000