水野 由子

© 2016 IEEE. It is a mystery that neural network composed of neurons with fluctuating characteristics can transmit information well reliably. In this paper, we show, in a simulation using a 9×9 2D mesh neural network, 9 to 1 multiplex communication is possible with 99% correct rate. Neurons are modeled by integrate and fire model without leak. Spikes spreads from transmitting neuron groups, propagated as spike waves, and received by receiving neurons. Then, the receiving neurons classify from which neuron group the spike waves come by back propagation neural network (BPN) method.

© 2015 IEEE. It's important for understanding brain intelligence to investigate how the signal/information is flown in neuronal network. We observed spike trains obtained by one-shot electrical stimulation with 8 × 8 multi-electrodes in cultured neuronal networks. Each electrode is considered to collect spikes from several neurons. We then constructed code flow maps as movies of the electrode array to observe the code flow especially of 1101 and 1011. To quantify the flow, we calculated the cross-correlations of the maximum direction of the codes with lengths less than 8. Normalized cross-correlations in the maximum direction were almost constant irrespective of code. Thus, the analysis suggested that the local codes for electrode flow maintained the code shape to some extent and conveyed information in the neural network. Then we made simulation of such code flow, and could estimate rough characteristics of neurons including refractory period and distribution of connection weights between neurons.