Image Recognition Using Spiking Neural Networks

Spiking neural networks (SNNs), a successor to today's artificial neural networks (ANNs) represent a more realistic model of biological neuron functionality and is more computationally efficient. This predisposes it for efficient real-time pattern recognition and object detection tasks. While neurons in conventional ANNs communicate using a constant output value, neurons in SNNs communicate using spikes that arc distributed in time. This functionality brings some problems in the process of encoding information into spike-like representation as well as in the SNN training. In this paper, we address some of these issues and introduced our ongoing work on SNN development. The proposed spiking multilayer perceptron and convolutional architectures were evaluated on the N-MNIST dataset for handwritten digit recognition task; the results show that the performance of the proposed solutions is comparable to the state-of-the-art and they even outperform some other related works under the comparison.