Today, ``neural networks'' play an important role in the field of image processing and pattern perception.
The ``neuron model'' which is used in neural networks is the model of nerve cells, and it enables computers to simulate the action of the nerve cells. The usual neuron model treats only their basic actions. Therefore, to become more similar to the nerve cells, the chaos that is found in actual nerve cells is introduced into the neuron. It is called ``chaotic neuron''.
The purpose of this study is to realize search access by chaotic neural networks in which the chaotic neuron is used.
The search access is that several patterns are learned in advance, and that a target pattern is searched for by inputing not its pattern itself, but its features.
So far, the search access has been realized by using only the special patterns with the high rate of success in searching. Therefore, in this study, I make random patterns, and examine whether the search access succeeds in searching by using them.
As a result, the search access succeeds in searching by using the patterns which meet at nearly right angles. But the rate of success is low, when they contain many patterns at right angles. It does not succeed in searching by using the other patterns. The reason considered is that they are not learned.
Therefore, the control by presynaptic inhibition is changed to raise the rate of success in searching. And, when the patterns can be learned, the search access by using the control can raise the rate of success.