Independent Component Analysis (ICA) and Blind Source Separation (BSS)

What is Independent Component Analysis?

Short introduction   Tutorial paper   Survey paper  

Introductory book "Independent Component Analysis"

NEW! Advanced book "Handbook of Blind Source Separation, Independent Component Analysis and Applications"

Current and recent research projects (2006-2010)

You can find summaries with references on our current and recent research projects related to independent component analysis (ICA) and blind source separation (BSS) here:

• Denoising source separation
• Nonlinear ICA and BSS


• Progress report on our ICA and BSS research in 2008-2009: Introduction, Non-negative projections, and Reconstruction of historical climate data by Gaussian-process factor analysis.
• Progress report on our ICA and BSS research in 2006-2007: Introduction, Convergence and finite-sample behaviour of the FastICA algorithm, Independent subspaces with decoupled dynamics, Extending ICA for two related data sets, ICA in CDMA communications, Non-negative projections, and Climate data analysis with DSS.
• Progress report on our Bayes research in 2006-2007 includes Nonlinear BSS and ICA, Nonlinear state-space models, and Non-negative blind source separation.

See also the Bayes Group and the Neuroinformatics Group

Free software developed by us

The FastICA software package implements in various environments the fixed-point algorithm. It is currently the most popular ICA algorithm because of its fast operation and applicability to large-scale problems.

Icasso: Software for investigating the reliability of ICA estimates by clustering and visualization

A Matlab package for denoising source separation, which is a semi-blind source separation technique based on denoising procedures.

Matlab packages for Nonlinear factor analysis (NFA) and Nonlinear dynamical factor analysis (NDFA). NFA can be used for nonlinear PCA and BSS, and NDFA is its extension for blind identification of a nonlinear dynamic state-space model.

A Matlab package for Projective nonnegative matrix factorization (PNMF).

Demos

Short demo      Try-it-yourself demo: Cocktail party problem  

Cocktailkutsu-demonstraatio sekoittuneiden äänten erottelusta  

People involved in ICA and BSS research:

• Erkki Oja (Professor, leader of the ICA group)
• Juha Karhunen (Professor, leader of the Bayes group)
• Antti Honkela (Post-doc researcher, Bayes group)
• Alexander Ilin (Post-doc researcher, Bayes group)
• Jaakko Luttinen (Researcher, Bayes group)
• Tapani Raiko (Post-doc researcher, Bayes group)
• Ricardo Vigario (Post-doc researcher, leader of the Neuroinformatics group)
• Zhirong Yang (Post-doc researcher)
• Zhijian Yuan (researcher)

Former researchers

Prof. Aapo Hyvärinen, Dr. Patrik Hoyer, Dr. Markus Harva, Dr. Ella Bingham, Prof. Jyrki Joutsensalo, Dr. Harri Valpola, Dr. Jaakko Särelä, Dr. Jarmo Hurri, Kimmo Kiviluoto, Dr. Mika Inki, Dr. Raju Karthikesh, Dr. Petteri Pajunen, Simona Malaroiu, Razvan Cristescu, Tomas Östman.

Earlier research projects (2000-2005)

• FastICA algorithm for independent component analysis
• ICA and its extensions as models of natural image statistics
• Icasso: software for stability analysis of independent components


• Progress report on our ICA and BSS research in 2004-2005: Introduction, Finite sample behaviour of the FastICA algorithm, Nonlinear ICA and BSS, Denoising source separation, Climate data analysis with DSS, ICA and denoising source separation in CDMA communications, ICA for image representations, Analyzing 0-1 data.
• Progress report on our Bayes research in 2004-2005 includes Nonlinear and non-negative blind source separation, and Dynamic modelling using nonlinear state-space models.
• Progress report on our ICA and BSS research in 2002-2003: Introduction, Theoretical advances, Comparison studies on blind separation of post-nonlinear mixtures, Text mining, ICA for astronomical data, ICA in CDMA communications, Explorative investigation of the reliability of independent component estimates, The European joint project BLISS.
• Progress report on our Bayes research in 2002-2003 includes Nonlinear static and dynamic blind source separation, Hierarchical modeling of variances, and Applications.
• Progress report on our application of ICA in 2000-2001: Decision trees using independent component analysis, ICA for text mining, ICA for analyzing financial time series, ICA for astronomical data, and ICA in CDMA communications.
• Progress report on our Bayes research in 2000-2001 includes Nonlinear factor analysis and independent component analysis, Nonlinear dynamic state-space models, and Applications.

ICA link collection

Click here for a collection of program packages, data sets and links