ICA Page
Data exploration, dimension reduction, separating mixed signals, automatic feature extraction - all are common techniques used across many domains of science in order to represent data so that it can be understood and easily managed or manipulated. Over the past few years, ICA has become an important tool for doing these tasks.
ICA SIG - An ICA SIG (Special Interest Group) has been formed for L.I.T. researchers- we meet approximately every two weeks, discuss papers, basics, applications, etc. Please see below if you would like to be on our mailing list or participate.
Please see the Graphical Models page
for information about this related are of interest.
Introduction to ICA
0) Necessary background
1) One-page
intro by Te-Won Lee
2) One-chapter
intro from Aapo Hyvärinen and Erkki Oja
ICA Links
ICA
Central (Cardoso)
Paris
Smaragdis' page full of links
Tony Bell's Salk
Institute, Computational Neurobiology Laboratory
Code Resources
MATLAB
and C code (Infomax from Bell and Sejnowski and more)
BLISS
Java and MATLAB code (including techniques for nonlinear mixtures)
FAST
ICA from the Finland ICA Mafia - The most popular quick-and-easy tool
for exploring ICA.
Applications
Text Mining, Face Recognition, Cocktail Party Phenomenon,
fmri separation, Noise Cancellation, Classification, MPEG-7 description,
Preprocessing for Neural Networks, Kitchen Sink washing.
PCA - Principle Components Analysis: an old and standard technique for data analysis and dimension reduction
Other basic statistical concepts:
Mean, Variance, Standard Deviation, higher-order
moments, cumulants.
Correlation, covariance, cross correlation, cross
variance.
SVD - Singular Value Decomposition - a basic matrix manipulation technique that has many purposes including computational linear system of equation solving, stability analysis, and dimension reduction.
The best place to find clear explanations of these things is in undergraduate
textbooks.
A on-line resource for math basics is: http://mathworld.wolfram.com/letters/
(see the subject index)
Numerical Recipes has a good discussion of SVD.
MATLAB is excellent for exploring, plotting, seeing how things work.
Contact : Lonce Wyse
email: lonce@lit.org.sg
phone: (x2010)