Training with Noise Addition in Neural Network Solution of Inverse Problems: Procedures for Selection of the Optimal Network⁎

Addition of noise to the patterns presented to a neural network during its training is a method to increase noise resilience of the trained neural network. However, the effect depends on the level of noise added. This article reports the first results of the study on elaboration of a procedure to select the optimal network or network subset for a given out-of-sample pattern from a set of networks trained with various noise levels, at the example of a model inverse problem.