High-dimensional model recovery from random sketched data by exploring intrinsic sparsity (Q782446)

This paper proposes randomized reduction methods to solve large-scale and high-dimensional machine learning problems, which can greatly speed up the modeling process by reducing either the dimensionality or the scale of the data. Furthermore, the authors theoretically show that the developed methods can recover well the optimal models built from the original data. This model recovery is achieved by using the intrinsic sparsity of optimal solutions and does not rely on any stringent assumption. Empirical results are also included to support both the method and the theory.