In this laboratory, the learning methods of a network have been studied based upon the Hebbian rule. Supplementary learning method proposed by Asakawa and improved by Kunieda has excellent learning efficiency. But this learning model is not followed with the Hebbian rule on detailed point.
In this study, the experiment of the supplementary learning method which was changed to follow the Hebbian rule are carried out. Conventional supplementary learning method uses external input to distinguish an unknown pattern from the known patterns, but the model changed use its own output. And when an unknown pattern is input into this network, the conventional model resets the inside. But the model changed can not be reset.
The learning method changed has low performance in learning and is weak against compared with the conventional learning method. Particularly, the function to reset the inside is important for the learning of a network. In this study, the learning method which follows the Hebbian rule is not always having high performance in learning.
協調