Wu Su

Sparsely connected rank-inspired neural network

A two-stage framework is developed, in which an approximately orthonormal neural basis is first constructed, and the output weights are subsequently determined using least-squares methods.