Regularized matrix inversion on a neural network architecture
Ronald J. Steriti, Michael A. Fiddy, J. Coleman · 1991
Summary form only given, as follows. A neural network architecture based on the Hopfield model has been studied which calculates the inverse of a matrix. An algorithm was then developed to simulate this architecture and tested for a known ill-conditioned matrix. This matrix inversion algorithm was also tested by using it in an image reconstruction algorithm and comparing it with the SVD inversion algorithm. The calculated inverses were examined closely, and the different pseudo-inverses were compared by calculating and comparing their singular value spectra. The relative merits of the different approaches were also compared.>