Structured Liquids in Liquid State Machines

Publication date

DOI

Document Type

Master Thesis

Collections

Open Access logo

License

CC-BY-NC-ND

Abstract

In this thesis we look at improving the performance of Liquid State Machines on a speech recognition and a music recognition task, by making structural changes to the liquid. The changes are designed, rather than evolved or learnt, to have more insight into what the effect of particular changes on performance are. We conclude that the effect of any designed structural changes is far outweighed by the random generation of liquids and that the evaluation of LSMs should focus on using sufficiently complex and preferably realistic tasks.

Keywords

LSM, Liquid State Machine, Music Recognition, Speech Recognition, Maass, Natschläger, Markram, Bach, Beethoven, Spiking Neural Networks, Supervised Learning, Time Series

Citation