Classification using spiking neural networks

Abstract

The functioning of the brain has attracted the interest of people since ancient times, how it processes the data has always became a subject of wonder and today, the mystery has still not been solved completely. The major developments in this area occurred in the last century with the enhanced knowledge of neuron functioning and development of more complicated and detailed neuron models. Consequently, the artificial neural networks which are designed as modeling the biological neurons become a significant area of Artificial Intelligence. Biological neurons transmit information by using sudden and short voltage increases called action potentials or spikes. Spiking Neural Networks are modeled just like biological neurons and use time information to fire these spikes. The most important features that distinguish Spiking Neural Networks - that are also called third generation neuron models - from the previous generations are processing the time information as spikes and keeping multiple connections in between consecutive neurons. Eventhough computational complexity and training time increases, spiking neural networks are very important because they have greater biological similarities and solve real world problems. The main purpose of this thesis is to clear that spiking neural networks can be used in classification problems, back-propagation algorithm can be used to train, in this way, real time datasets can be classified successfully.