Information processing using liquid state machines based on spiking neurons

Abstract

Spiking neural networks are introduced as the third generation of neural models. They are dynamic models that potentially have much more processing power than classic neural networks. This thesis presents a novel approach to perform Gabor filtering using Liquid State Machines based on Spiking Neurons. The Liquid State Machine is a powerful architecture that is capable of performing universal computations without being trained on specific data. It is the job of special readout units to interpret the computation results and map them on specific target functions. In addition, the thesis presents tools that allow fast simulating of large neural networks by running the simulation in parallel on a GPU.