内部記憶を持つニューラルネットワークにおける遅れ学習と生成された内部記憶

概要

論文の詳細を見る
Elman presented a network with a context layer for the time-series processing. The context layer is connected to the hidden layer for the next calculation of the time series, which keeps the output of the hidden layer. We reformed the context layer into the internal memory layer, which is connected from the hidden layer with the connection weights to make the internal memory. Then, the internal memory is not a simple copy of the hidden layer. We also changed the "back-propagation through time" to the time-delayed back-propagation learning to generate the internal memory. In this paper, the ability of the network with the internal memory layer is demonstrated by applying it to the simple sinusoidal discrete-time time-series, and the process of generating the internal memory is investigated with 2-dimentional plots.
岐阜工業高等専門学校の論文