Bounding the performance of neural network estimators, given only a set of training data

摘要

Uses a neural network method for obtaining a stochastic Cramer-Rao bound on estimates, given only the training data. The Cramer-Rao bounds can be used (1) to help determine when neural net training should be stopped, (2) to re-order the network inputs according to their contributions to the bounds, and (3) to eliminate less useful inputs. The convergence of the modelling procedure is shown. Examples are provided to illustrate the method.