N3-CPL: Neuroplasticity-based neuromorphic network cell proliferation learning

Abstract

In general, Spiking Neural Networks (SNNs) have a network structure with special methods applied to neuron models and information transmission to mimic humans biologically. However, the existing SNN structures have two problems, such as fixed existing Artificial Neural Network structures and difficulty in learning due to lack of spike information during information transfer. Recently, many approaches of learning SNNs have been proposed in order to alleviate those two drawbacks. However, it is very difficult to overcome the drawbacks only by the learning method without the fundamental solution for the structure. In order to solve the problem of structure and learning method, we propose a novel flexible network construction method using neurogenesis-based cell proliferation concept and Triple Simultaneous- Spike Timing Dependent Plasticity (TS-STDP) which is improved learning method through neuroplasticity-based spike timing. We build the network flexibly and automatically by employing the concept that not only one neuron exists in the neural network, but also that various cells proliferate and transform from stem cells to function. In addition, TS-STDP is designed by considering the correlation of signal response among several neurons to solve the lack of information due to spike sparsity, which is a disadvantage of STDP. In the experimental section, we demonstrate and analyze our method using Mixed National Institute of Standards and Technology image data. Our method is 2.7× better in memory efficiency and 1.7× better in computational efficiency than the existing method. In particular, the research that automatically constructs the network structure is the first to my knowledge.