Today, neural network which can work simultaneously attracts a great deal of attention as anti-Neumann-computer.
Neural network has some types. The associative memory model is one of them. It can raise its remembering ability by putting outputs back to itself. The model is studied to increase its ability. Two years ago, Mr. Yamabe who was a research worker in this office had proposed the model that can keep previous state (multi-layer model), and he confirmed its ability.
In this paper, I proposed the random-layer model which applied feedback mentioned above, and measured its ability. This model was made according to different thoughts of the multi-layer model. I compared this model with the multi-layer model. As a result, I confirmed that it can correct errors about 25%. Besides, this model can remember the origin pattern in case learning patterns are in different cycles. So I confirmed that regardless of the length of recurrence part series, the random-layer model will be sufficient if there are two layers in this model. Moreover, I confirmed that its remembering ability is not influenced even if it cannot know the beginning pattern.