Static Pattern Recognition using Liquid State Machines
Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
The Liquid State Machine is a computational model introduced by Wolfgang Maass. Based on a short term memory model, it is a promising step in mimicking the adaptability and efficiency of the learning process found in neural cortices of living organisms. It has been successfully implemented using articial Spiking Neural Networks.
In this report we investigate the aforementioned adaptability and specically compare performance between two problems reusing essential parts of the network. Using publicly accessible tools and a simplistic approach we implement a Liquid State Machine for two static pattern recognition tasks: A basic XOR gate and classication of the Iris data set.
The fraction of correct answers of our Liquid State Machine reached a percentage of above 90% for both problems easily. A conclusion could not be drawn regarding the correlation of performance between the two problems. The inconclusiveness of the investigation was assessed to be caused mainly by the simplicity of both the benchmark problems, and possibly by the implementation of our Liquid State Machine.
Place, publisher, year, edition, pages
IdentifiersURN: urn:nbn:se:kth:diva-168133OAI: oai:DiVA.org:kth-168133DiVA: diva2:814442
Herman, Pawel, Assistant Professor