Independent component analysis. Theory and applications. (English) Zbl 0910.94004

Boston: Kluwer Academic Publishers. xxxiii, 209 p. Dfl 270.00; $ 118.00; £80.25 (1998).
This is a book devoted to a recently emerging domain: Independent component analysis (ICA). Loosely speaking, the goal pursued by ICA is to extract independent sources which are mixed in the global signal whose samples are recorded (observed). The tools which are used in the ICA approach are various; Bayesian statistics, Information theory, Neural networks, Adaptive filtering, etc.
To get an idea of the topics included, we list the contents: Introduction; Part I, (1) Basics; (2) Independent component analysis; (3) A unifying information-theoretic framework for ICA; (4) Blind separation of time-delayed and convolved sources; (5) ICA using overcomplete representation; (6) First steps towards nonlinear ICA; (7) Biomedical applications of ICA; (8) ICA for feature extraction; (9) Unsupervised classification with ICA mixture models; (10) Conclusions and future research; Bibliography; About the author; Index.
The interest for the ICA approach stems from the spectacular successes in blind deconvolution, analysis of electroencephalographic signals, extraction of features of images. The book provides an excellent introduction to ICA. Due to many examples of application and to the experience of the author, this volume is of great use for theorists as well for practitioners. It is worth pointing out that the ICA approach gives excellent results in many cases where the underlying hypotheses are not fulfilled. Therefore, there is still much to be done in the ICA domain.


94A12 Signal theory (characterization, reconstruction, filtering, etc.)
94-02 Research exposition (monographs, survey articles) pertaining to information and communication theory