ニューラルネットの入出力関係に関する一考察

概要

論文の詳細を見る
In this paper, the precision of function which expresses the relation between input and output of a system and is generated from a neural network is compared with that of the function generated from an identification model which is expressed b y a polynominal, and the effect of observation noise on the prescision is investi-gated by numerical simulation. The results obtained are summarized as follows. (1) In the generation of a sample function expressed by a square polinominal, the precision of function generated from the neural network is better than that of function generated from the identification model. (2) In the case of less than 2% of signal to noise ratio, a learning process which means the decreasing of the error between the sample function and generated functions with the increasing of the number of teaching values is realized in both neural network and identification model. (3) The precision of functions generated from both neural network and identification model grows worse rapidly with the increasing of signal to noise ratio.
岡山理科大学の論文