High-resolution image reconstruction based on matrix inversion on a fully connected architecture
Ronald J. Steriti, J Coleman, Michael A. Fiddy · Inverse Problems · 1990
The authors discuss the implementation of a technique first proposed by Jang et al. (1988) for the inversion of a matrix. The inversion is performed by mapping an appropriate energy function onto a fully connected processing architecture (which is similar to a Hopfield neural network). They describe the advantages and disadvantages of inverting a matrix in this fashion by comparison with more conventional approaches. They also consider the use of the calculated inverse in a specific image reconstruction procedure. They conclude that there are advantages in inverting a matrix in this fashion even without the ideal parallel hardware and synchronous updating of a fully connected network; convergence of the process is reliable and can be very fast.