An artificial neural network model for generating periodic signals by synchronizing external stimuli

A living body has self-organized information processing systems in their nerve system. They construct various internal models of the environment autonomically. Periodic behaviors, such as circadian rhythm, locomotion of limbs in walking and tapping, come from these internal pattern generators. They obtain the periodic signal by use of external stimuli, that is, brightness of day lights and images from eyes. In this study, the authors represent that a model based on multilayer type neural network with feedback connections. This neural network can construct periodic pattern generators by training using external input signal. Arbitrary periodic signal can be generated if the network trained properly. Both periodic scalar signals and periodic image signals were examined to investigate characteristic of the proposed model. After the neural network training, they generated periodic signals that are identical to training pattern autonomically. In addition, stability of the generation of periodic signals was observed even if noise was mixed in the signal generation process. This characteristic is corresponding to homeostasis of biological system. These results show a pattern generator similar to a biological system was constructed in the proposed model. This neural network model is expected to be convenient as a model to analyze the fundamental mechanism of the brain function.