Heterogeneous Hybrid Neural Network for Modeling Spatially Distributed Destructive Processes
Volodymyr Sherstjuk, Maryna Zharikova · 2019
In this paper, a hybrid model of cellular neural network based on het- erogeneous spiking neurons connected with a spatial model of terrain through hierarchically ordered context is proposed. Spatial cells are considered as ele- ments of the neural network (neurons) and events are considered as spikes at the neuron output. A discrete automaton model with integrated likelihood model sup- plements a hybrid spiking neuron model to determine the neuron state at specified time points based on probability or possibility of state transitions. The structure of neuron connections in the model resembles a cellular network model. The hi- erarchical context containing a set of transmitters allows organizing additional channels of communication between neurons and provide remote sensing data to the neural network. Neurons have controlled sensitivity to transmitters with spe- cial receptors connected to the network context. The method of modeling spatial distributed destructive processes using the proposed hybrid neural networks is presented. The proposed method and models are intended for modeling dynamic systems with different types of simultaneously arising interacting processes with respect to their spatiotemporal aspects.