Neural network processing as a tool for function optimization

Abstract

We summarize our recent work on the development of ‘‘neural network’’‐like processing for function optimization, and demonstrate how this method can be designed so as to avoid trapping in local extrema. We illustrate the application of our algorithm by considering the inversion of severely ill‐posed remote sensing data and the solution of variational problems. The algorithm described here has been implemented on a serial processor, but is cast in a form which is ideally suited for parallel processing.