Variable Binding Through Assemblies in Spiking Neural Networks

Research output: ThesisMaster's ThesisResearch

Abstract

A thorough understanding of the processes underlying the cognitive capabilities of humans has remained elusive. Among the open issues is the binding problem, i.e. the question of how bits of information are tied together in the brain. This work tackles the problem using computer simulations which show that networks of spiking neurons can perform simple binding operations through their dynamics. The circuit is built on mechanism used in the brain which have been firmly established by experimental studies: winner-take-all dynamics within groups of neurons as well as spike-timing dependent plasticity. The model in this work is based on an existing model, which is improved with respect to the biological plausibility of the implementation. A new mode of information storage is introduced which is consistent with findings of experimental neuroscience concerning the retention of information in working memory in the human cortex. Furthermore, an optimization algorithm with very few hyperparameters is introduced which allows the rapid optimization of high-dimensional parameter spaces with constraints. This algorithm was used to tune the parameters of the variable binding model.
Original languageEnglish
QualificationMaster of Science
Awarding Institution
  • Graz University of Technology (90000)
Publication statusPublished - 2017

Cite this

Variable Binding Through Assemblies in Spiking Neural Networks. / Müller, Michael Günther.

2017.

Research output: ThesisMaster's ThesisResearch

Müller, MG 2017, 'Variable Binding Through Assemblies in Spiking Neural Networks', Master of Science, Graz University of Technology (90000).
@phdthesis{e155998c77f84531b3a435155aee9e02,
title = "Variable Binding Through Assemblies in Spiking Neural Networks",
abstract = "A thorough understanding of the processes underlying the cognitive capabilities of humans has remained elusive. Among the open issues is the binding problem, i.e. the question of how bits of information are tied together in the brain. This work tackles the problem using computer simulations which show that networks of spiking neurons can perform simple binding operations through their dynamics. The circuit is built on mechanism used in the brain which have been firmly established by experimental studies: winner-take-all dynamics within groups of neurons as well as spike-timing dependent plasticity. The model in this work is based on an existing model, which is improved with respect to the biological plausibility of the implementation. A new mode of information storage is introduced which is consistent with findings of experimental neuroscience concerning the retention of information in working memory in the human cortex. Furthermore, an optimization algorithm with very few hyperparameters is introduced which allows the rapid optimization of high-dimensional parameter spaces with constraints. This algorithm was used to tune the parameters of the variable binding model.",
author = "M{\"u}ller, {Michael G{\"u}nther}",
year = "2017",
language = "English",
school = "Graz University of Technology (90000)",

}

TY - THES

T1 - Variable Binding Through Assemblies in Spiking Neural Networks

AU - Müller, Michael Günther

PY - 2017

Y1 - 2017

N2 - A thorough understanding of the processes underlying the cognitive capabilities of humans has remained elusive. Among the open issues is the binding problem, i.e. the question of how bits of information are tied together in the brain. This work tackles the problem using computer simulations which show that networks of spiking neurons can perform simple binding operations through their dynamics. The circuit is built on mechanism used in the brain which have been firmly established by experimental studies: winner-take-all dynamics within groups of neurons as well as spike-timing dependent plasticity. The model in this work is based on an existing model, which is improved with respect to the biological plausibility of the implementation. A new mode of information storage is introduced which is consistent with findings of experimental neuroscience concerning the retention of information in working memory in the human cortex. Furthermore, an optimization algorithm with very few hyperparameters is introduced which allows the rapid optimization of high-dimensional parameter spaces with constraints. This algorithm was used to tune the parameters of the variable binding model.

AB - A thorough understanding of the processes underlying the cognitive capabilities of humans has remained elusive. Among the open issues is the binding problem, i.e. the question of how bits of information are tied together in the brain. This work tackles the problem using computer simulations which show that networks of spiking neurons can perform simple binding operations through their dynamics. The circuit is built on mechanism used in the brain which have been firmly established by experimental studies: winner-take-all dynamics within groups of neurons as well as spike-timing dependent plasticity. The model in this work is based on an existing model, which is improved with respect to the biological plausibility of the implementation. A new mode of information storage is introduced which is consistent with findings of experimental neuroscience concerning the retention of information in working memory in the human cortex. Furthermore, an optimization algorithm with very few hyperparameters is introduced which allows the rapid optimization of high-dimensional parameter spaces with constraints. This algorithm was used to tune the parameters of the variable binding model.

M3 - Master's Thesis

ER -