Details
The Independent Component Analysis (ICA) is a matrix factorization method that utilizes Gram-Schmidt orthogonalization by considering two vectors (two signals) to be orthogonal if their difference is Gaussian white noise.
The argument matrix of can be a square or rectangular matrix.
The columns of the argument matrix are interpreted as signals.
Non-negative matrix factorization (NNMF) is a matrix factorization method that reduces the dimensionality of a given matrix.
NNMF has two factors: the left one is the reduced dimensionality representation matrix and the right one is the corresponding new-basis matrix.
The argument matrix need not be square.
NNMF allows easier interpretation of extracted topics in the framework of latent semantic analysis.
When using NNMF over collections of images and documents, often more than one execution is required in order to build confidence in the topics interpretation.
In comparison with the thin singular value decomposition, NNMF provides a non-orthogonal basis with vectors that have non-negative coordinates.