From Bacteria, a Consideration of the Evolution of Neural Network

The neural network proposed in this paper hierarchically processes time series data and has the function which becomes the basis of intellectual behavior of not only bacteria but also evolved animals. At first, time series data are divided into sequences of subsequences in which the same element does not appear multiple times. The neural network that recognizes/generates the obtained subsequence is familiar to basic behavior of nerve cells and is called a Basic Unit. General time series data have a hierarchical structure. The lowest level is the sequence of the subsequence obtained as a result of division. A neural network composed of hierarchically connected Basic Units processes time series data. Hierarchical processing is performed according to the context structure of time series data. The place where the Basic Units are activated moves from upper layer to lower layer or in the opposite direction as processing progresses. It is possible to predict the next processing by using the contextual position of the current executing process. There is a plurality of neural networks which process time series data according to the category of time series data. Isomorphism between neural networks brings about isomorphism of context structure of processing process. The behavior of mirror neurons is explained using the interaction between isomorphic neural networks.