Dimensionality reduction using SVD Use only dimensions Retain first columns for and and first values for First columns of give the basis vectors in reduced space Best rank approximation in terms of sum squared error