Learning in clustered spiking networks
File(s)
Author(s)
Maes, Amadeus
Type
Thesis
Abstract
Neurons spike on a millisecond time scale while behaviour typically spans hundreds of milliseconds to seconds and longer. Neurons have to bridge this time gap when computing and learning behaviours of interest. Recent computational work has shown that neural circuits can bridge this time gap when connected in specific ways. Moreover, the connectivity patterns can develop using plasticity rules typically considered to be biologically plausible. In this thesis, we focus on one type of connectivity where excitatory neurons are grouped in clusters. Strong recurrent connectivity within the clusters reverberates the activity and prolongs the time scales in the network. This way, the clusters of neurons become the basic functional units of the circuit, in line with an increasing number of experimental studies. We study a general architecture where plastic synapses connect the clustered network to a read-out network. We demonstrate the usefulness of this architecture for two different problems: 1) learning and replaying sequences; 2) learning statistical structure. The time scales in both problems range from hundreds of milliseconds to seconds and we address the problems through simulation and analysis of spiking networks. We show that the clustered organization circumvents the need for non-bio-plausible mathematical optimizations and instead allows the use of unsupervised spike-timing-dependent plasticity rules. Additionally, we make qualitative links to experimental findings and predictions for both problems studied. Finally, we speculate about future directions that could extend upon our findings.
Version
Open Access
Date Issued
2021-09
Date Awarded
2022-01
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Clopath, Claudia
Barahona, Mauricio
Sponsor
EPSRC Centre for Doctoral Training in Neurotechnology for Life and Health
Publisher Department
Bioengineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)