Local independent component analysis
David G. Grier, Jack D. Cowan, Juan K. Lin · 1997
The study of statistical independence is of fundamental importance to the field of statistics and of all empirical sciences. In scientific research, one often assumes the variables of interest are statistically independent in order to proceed with the analysis. On other occasions, one wishes to find revealing relationships between the variables. From the observables, analysis decomposes the observables into underlying causes, while synthesis proceeds in the opposite direction. Like the principle of superposition in physics, investigation begins with the simplest linear dependence between an observable and the underlying causes. This is the subject of linear independent component analysis. Improvement over current approaches as well as various generalizations are the topic of this thesis.