Study on Propagation Efficiency and Fidelity of Subthreshold Signal in Feed-Forward Hybrid Neural Network under Electromagnetic Radiation
Wang Guowei,Ge Mengyan,Lu Lulu,Jia Ya,Zhao Yunjie
DOI: https://doi.org/10.1007/s11071-021-06247-z
IF: 5.741
2021-01-01
Nonlinear Dynamics
Abstract:The hybrid neural model provides a computationally effective and biophysics-based neuron behavior model, which maintains its simplicity by incorporating the dynamics characteristics of ion channels in two fast ion channels. In view of the issue of signal propagation in multilayer network composed of hybrid neurons is still unclear and the role of neural network is mainly used to generate biological consciousness and help creatures to think and act, it is necessary to explore neural network with more perfect function and better transmission efficiency. Considering a feed-forward neural network (FFN) which is composed of hybrid neurons in the presence of electromagnetic radiation, the effects of the Gaussian white noise, the strength of synaptic weight and inter-layer connection probability on subthreshold excitatory postsynaptic current (EPSC) propagation are investigated. In order to clarify the mechanism of signal transmission more clearly, the dot raster plot of spike and the corresponding post-spike time histogram of each layer are explored. Particularly, the propagation efficiency and fidelity of subthreshold EPSC signal in the FFN are investigated by calculating the power norm and the spike timing precision. Our results indicate that both the power norm and the spike timing precision fluctuate with the increase in the inter-layer connection probability, and the multi-periods bursting discharge mode can be detected in both the transmission layers and the output layer. Besides, there exists an optimal noise intensity to guarantee the best temporal coding and maximum transmission efficiency of subthreshold EPSC signal. Moreover, the optimal noise intensity makes the fidelity of the FFN the highest, which ensures that the weak signal and the excitation caused by noise can be distinguished. The conclusions obtained in this paper have potential value for discussing the encoding, decoding and propagation mechanism of information in real neural networks.