Regularized hopfield neural networks and its application to one-dimensional inverse problem of magnetotelluric observations
Yuanchou Zhang, Ken V. Paulson · Inverse problems in engineering · 1997
Hopfield neural networks are massively parallel automata that support specific models and are adept at solving optimization problems. However, the standard Hopfield neural network approach suffers from a “rough” solution space and convergence properties that are highly dependent on the starting model. Globally-optimum solutions are not guaranteed. These drawbacks may be overcome by introducing regularization into the network in the form of local feedback smoothing. Application of regularized Hopfield networks to over fifty optimization test cases have yielded successful results, even with uniform (minimal information) starting model. The non-linear, one-dimensional magnetotelluric inverse problem has been solved by means of the regularized network. The method compares favorbaly with other techniques.