- Method to extract independent sources given only the observed
data that are a mixture of the unknown sources
- Like PCA, ICA also decorrelates the signals (2nd order
statistics)
- ICA also reduces higher-order statistical dependencies-attempting
to make the signals as independent as possible
- Definition: ICA is a way of finding linear non-orthognal
co-ordinate system in any multivariate data
- ICA approach tries to maximize the mutual information between
the assumed input sources and the observed output data vectors
- ICA has been more succesful from a Neural Network base approach