ILS-based Reservoir Computing for Handwritten Digits Recognition

Abstract

ILS-based reservoir is a bio-inspired computational model consisting of spiking neurons which has been designed to process spatiotemporal patterns appropriately. In ILS-based reservoir, the neurons are located in an ionic environment and the connections are provided by ionic density. By using ionic diffusion as a processing operation, this model is able to consider the effects of both the preceding and current stimuli properly. Since character recognition is an important task in various applications, this paper focuses on the classification of handwritten digits using ILS-based reservoir. For this purpose, a neuromorphic handwritten digit dataset called N - MNIST dataset is used as a benchmark. Like all reservoir networks, a readout layer is added to the ILS-based reservoir for classifying this dataset. Classification of the N-MNIST using the proposed network model has resulted in a maximum accuracy of 97.69 % which is comparable to state-of-the-art works using spiking neural networks.