出力層での誤差の自乗和の値によって変化する学習係数を持つ階層ニューラルネットワークの学習アルゴリズム

概要

論文の詳細を見る
In this paper, we propose a fast learning algorithm for the multi-layered neural network which changes the learning coefficient for the hidden units (used to a fixed parameter in general) by the sum of the square error for the output units. The algorithm is able to accelerate learning by selecting appropriately the ratio of the learning accerelation coefficient to the learning continuation one defined in the paper. The proposed algorithm is an extention of conventional back-propagation learning algorithm. This learning algorithm makes more acceleration possible of learning, one of weak points on back-propagation. The effectiveness of this method on convergence of the sum of the square error for the output units has been verified by the computer simulation for two examples of an exclusive-or problem and an encoding one using the proposed learning algorithm.
東海大学の論文