Improved perception-based spiking neuron learning rule for real-time user authentication

Abstract

Spiking neural networks (SNNs) have been highly successful in spatiotemporal pattern recognition. As one of the most efficient supervised learning algorithms in spike sequences learning, the perceptron-based spiking neuron learning rule (PBSNLR) still has a relatively high computational complexity, which is difficult to use in a real-time system. In this paper, a novel method is presented to improve PBSNLR׳s efficiency without reducing its accuracy, and this method is applied to solve user authentication problem in real time. In our method, a user׳s behavioral biometric of sliding dynamic and finger pressure are selected as spatiotemporal features to recognize the user׳s identity. The temporal feature is obtained by the time coding of SNNs and the spatial feature is represented by the neurons׳ relative positions. Comprehensive experimental results demonstrate that our improved algorithm outperforms the traditional PBSNLR in terms of efficiency and exhibits excellent performance when identifying users of touch screen devices.