next up previous contents
Next: 目次

概要:

  When the neural network does the associative memory, it is more efficient to use incremental learning than correlative learning.

The network might forget information that has already been remembered when it keeps learning by incremental learning for a long time. When the network has much information to remember, it is likely to forget information that has been remembered. However, even if some of the remembered information are lost, the contents that was once remembered will never be useless. Because the total amount of information will increase over the entire network.

Forgetting is due to refractoriness. But as long as there is less information to be remembered, refractoriness does not influence the study of the network. It is necessary to provide strength of the refractoriness carefully when a lot of information has to be memorized.





出口研究室へ

Toshinori DEGUCHI
2005年 4月 1日 金曜日 17時24分52秒 JST