
Separating mixed signals into statistically independent sources.Independent Component Analysis (ICA) is a computational technique for separating a multivariate signal into additive subcomponents, assuming the subcomponents are non-Gaussian and statistically independent from each other. It is best known for solving the 'cocktail party problem': extracting a single voice from a recording of multiple overlapping conversations. ICA is a cornerstone of blind source separation, where neither the sources nor the mixing process are known beforehand.
ICA was formally introduced in the 1980s and 1990s, with key contributions by Pierre Comon (1994) who defined the problem and by Aapo Hyvärinen and Erkki Oja who developed the FastICA algorithm in 1997.
ICA models the observed signals as a linear mixture of unknown independent sources. The algorithm searches for a linear transformation that maximizes the statistical independence of the resulting components. Independence is typically measured by non-Gaussianity, since mixtures of independent variables tend toward Gaussian distributions (by the central limit theorem). ICA iteratively adjusts the transformation to maximize a measure like kurtosis or negentropy, thereby recovering the original sources up to scaling and permutation.
ICA is used in brain imaging (EEG and fMRI) to isolate neural signals from artifacts, in finance to identify hidden factors in stock returns, and in audio processing to separate mixed recordings.
ICA can separate two people speaking simultaneously in a room using only two microphones, even though the microphones each pick up a blend of both voices.





