Building a self-learning memristor-based spiking neural network on handwritten digit recognition and orientation extraction

In this paper, we present a memristor-based spiking neural network to identify handwritten digit figure and extract orientation. Digit figures are from MNIST database. Orientation spikes are generated in response to relative changes in illumination at the pixel level and transmitted to the spiking neural network. The network is a two-layered structure consisting of integrate-and-fire neurons and memristor. Memristors are used as synapses in this neural network performing learning. The memristors learn through an adaptation of spike-time-dependent plasticity (STDP) for training. Spiking neurons are arranged in a winner-take-all (WTA) circuit, which is one of the most frequently studied connectivity patterns. Neurons become sensitive to different patterns of pixels through learning. The simulation results illustrate memristors perform well on unsupervised learning and the biologically inspired learning scheme is capable of generating selective to different patterns on digit recognition and orientation extraction. ? 2014 WIT Press.EI