COMPLEX DATA CLUSTERING WITH SINGLE-LAYER DYNAMICALLY LINKED SPIKING NEURAL NETWORK

Abstract

An improved method for constructing and training single-layered spiking neural networks is preposed. The method allows to apply single-layered spiking neural networks that encode each data dimension by one neuron of the input layer for the recognition of complex and overlapping clusters with procedure of unsupervised training. The presented approach allows to obtain acceptable accuracy of the classification with the ability to detect complex data clusters at considerably simplified structure of the neural network.

Keywords

• импульсные нейронные сети
• классификация данных
• хеббовское обучение
• обучение “без учителя”.