Reduced Basis Decomposition

Employing a greedy algorithm, RBD approximates the column space of a matrix X faster than SVD does.

Reduced Basis Decomposition (RBD) is a new decomposition strategy inspired by the Reduced Basis Method. Given X the high-dimensional data, RBD approximates it by a product YT with Y being the low-dimensional surrogate and T the transformation matrix. This is done through a greedy algorithm thus very efficient. In fact, it is significantly faster than SVD with comparable accuracy in many cases shown.
RBD has been demonstrated to work well as a dimension reduction tool for many data mining tasks.