Improving Spiking Neural Network Performance Using Astrocyte Feedback for Farsi Digit Recognition

The third generation of neural networks is called spiking neural networks. Spiking neural networks can not only answer all the problems that can be solved by common neural networks, they can also be computationally more powerful than them. Also, these networks are closer to the biological reality of neurons. According to these reasons, spiking neural networks have received much attention in recent years. In addition to neurons, other elements are also used in spiking neural networks, whose connection with neurons has been proven biologically. These elements are called astrocytes. In this study, we have used astrocyte local feedback in a spiking neural network to improve the accuracy of the network in a digit recognition process. To reach this goal, we use the network proposed by Diehl et al in 2015, then we added astrocyte local feedback to this network. We run the network in both states. The classification accuracy of the network in the absence of astrocytes was 88.48% on Farsi handwritten digits. After adding astrocyte local feedback, it is observed that astrocytes control the activity of neurons and modified the weight of connections based on neuron activity. In this state classification accuracy reached 90.26%. The idea was to create a neural network for learning, where, inspired by the physiological role of astrocytes at the tripartite synapse improves the performance of the neural network and we demonstrate that systems that include neurons and astrocytes perform better than systems incorporating only neurons.