Supervised Learning with Small Training Set for Gesture Recognition by Spiking Neural Networks

Abstract

This paper proposes a novel supervised learning algorithm for spiking neural networks. The algorithm combines Hebbian learning and least mean squares method and it works well for small training datasets and short training cycles. The proposed method is applied in human-robot interaction for recognizing musical hand gestures based on the work of Zoltán Kodaly. The MNIST dataset is also used as a benchmark test tó verify the proposed algorithm's capability to outperform shallow ANN architectures. Experiments with the robot also provided promising results by recognizing the human hand signs correctly.