学位論文

ディジタル型誤差逆伝播法における学習改良の研究

jpn

2013-03-24

Not Applicable (or Unknown)

2013-03-24

修士(工学)

工学研究科情報電子工学専攻; 指導教員: 李嘉

A learning based error back propagation algorithm, a proposed non-differential digital back propagation by using two-valued output element is a multilayer neural network (DBP).Traditional simple Perceptron learning without teacher signal only to the output layer, DBP learning algorithm gives the teacher signal to reduce output error to the middle layer. In this paper, Non-linear separable issues to solve that Using the concept of tolerance for giving teacher signal in DBP algorithm, For Improving learning according to effect of an appropriate learning parameter settings, and the high speed calculation method by batch learning error function gradient Resilient back-propagation algorithm (Rprop algorithm), confirmed that fluctuations in the success rate and the number of learning.

Thesis