Investigation of Multi-Layer Perceptron with Pulse Glial Chain Including Neurogenesis

In this study, we propose Multi-Layer Perceptron (MLP) with pulse glial chain including neurogenesis. In this network, we one-by-one connect glia unit with neurons in a hidden-layer. The glia generates a pulse according to the connecting neuron output. This pulse increases the threshold of the neuron and excites the neighboring glias. The pulse generation frequency is also changed according to the connecting neuron output. Moreover, we introduce a neurogenesis to the network. By the neurogenesis, any neurons are removed and newborn neurons are set to same position. The removed neurons are chosen by the number of pulse generation of the glia, and the newborn neuron has random value in the connection with other neurons. We consider that the unimportant neurons for the learning are removed and the newborn neurons become the important neurons. By computer simulations, we confirm that the proposed MLP obtains a better solving ability than the previous MLP.