Construction And Analysis Of Cascaded Model Based On Dendritic Morphologic Neural Network

Brain-inspired computing(BIC)is a computing paradigm based on the structure of the human brain and information processing mechanism,and its construction refers to the basic composition of the brain.Most brain-inspired computing frameworks are based on neurons as the basic unit.Large-scale neuron networks are formed by coupling between neurons to achieve different levels of functions.Therefore,the construction of neuron models is one of the cores of brain-like computing.The biophysical model of neuron is too complex in neural computation,while the phenomenological model is not enough in physiologic interpretability,so how to construct a neuron and network model with both neuron geometry and biological firing mechanism is an important problem.In this paper,we propose a hierarchical generalized linear model(hGLM)to construct a neural model with dendritic morphology.Through analyzing the excitatory and inhibitory properties of neural pathway model,we study its physiological interpretability and construct neural network based on this cascaded model.Field prgrammable gate array(FPGA)is used to realize the network to prove the advantages of the cascaded model.Firstly,the neuron with dendritic morphology and neural network construction method are studied.Based on the information processing of single-neuron dendrite network,a neuron node with multi-compartment structure is constructed by hGLM.Based on the physiological connection mode between neurons,the cascaded channels with different communication modes are built between nodes,forming the cascaded model of neural network with physiological interpretability.Secondly,the cascaded model of neural network is constructed and its physiological characteristics are analyzed.By changing the topological structure of dendritic network and adjusting the excitation-inhibition ratio of dendrites,we find that the neural pathway model has the characteristics of neuron saturation,which confirms the effectiveness of neural pathway model to simulate the real nervous system.Furthermore,a neural pathway model of visual cortex based on hGLM is established to explore the effects of stimulus intensity,connection location and time delay on neuron firing activity,which verify the potential application value of this method in physiological experimental research.Finally,the computational advantages of neural network cascaded model is studied by hardware implementation.Based on hGLM,the FPGA framework and cascaded model of single-layer and double-layer neurons are designed,and the firing function of neurons which are implemented on hardware platform is verified according to the software simulation results.The realization of modular reconstruction from single neuron to neural network proves that the cascaded model of neural network has advantages in scalability and computational efficiency.The neural network cascaded model proposed in this paper describes the multiplexing information processing mechanism of neuron,which is a methodological contribution to the construction of physiologically explicable brain-inspired computing platform.The proposed hGLM-based neural network cascaded model provides technical support for the realization of efficient and scalable brain-inspired computing digital circuits.The results of this study are of great significance for the coding of spatiotemporal dynamics of nervous system in brain-inspired computing.