NEURAL NETWORK FOR SINGULAR VALUE DECOMPOSITION

A new massively parallel algorithm for singular value decomposition (SVD) has been proposed. To implement this algorithm an analogue neuron-like multilayer architecture with continuous-time learning rules has been developed. Extensive computer simulation experiments have confirmed the validity and high performance of the proposed algorithm. The proposed neural network associated with learning rules may be viewed as a nonlinear control feedback-loop system. This conceptual viewpoint enables many powerful techniques and methods developed in control and system theory to be employed to improve the convergence of the learning algorithm.