A selective learning algorithm for certain types of learningfailure in multi-layer perceptrons

摘要

Summary form only given, as follows. A simple selective learningnalgorithm for use with multilayer perceptrons is presented. Thisnalgorithm has proved useful in certain types of problems where learningnfailure occurs using standard backpropagation. Examples of thesenproblems are included. The algorithm is based on the RMS output errorncomputed across all output nodes and all training patterns. The learningnrate is decreased for all individual output nodes each time the error isnless than a user-chosen multiple of the RMS error corresponding to thenprevious pass. This algorithm has produced convergence where thenstandard fixed-again backpropagation failed