guest ::
| Home > Publications database > Simulation and theory of large-scale cortical networks |
| Home > Publications database > Simulation and theory of large-scale cortical networks |
| Book/Dissertation / PhD Thesis | FZJ-2023-03187 |
2023
Forschungszentrum Jülich GmbH Zentralbibliothek, Verlag
Jülich
ISBN: 978-3-95806-708-0
This record in other databases:
Please use a persistent id in citations: urn:nbn:de:0001-20231004081300012-6971752-0 doi:10.34734/FZJ-2023-03187
Abstract: Cerebral cortex is composed of intricate networks of neurons. These neuronal networks are strongly interconnected: every neuron receives, on average, input from thousands or more presynaptic neurons. In fact, to support such a number of connections, a majority of the volume inthe cortical gray matter is filled by axons and dendrites. Besides the networks, neurons themselves are also highly complex. They possess an elaborate spatial structure and support various types of active processes and nonlinearities. In the face of such complexity, it seems necessary to abstract away some of the details and to investigate simplified models.In this thesis, such simplified models of neuronal networks are examined on varying levels of abstraction. Neurons are modeled as point neurons, both rate-based and spike-based, and networks are modeled as block-structured random networks. Crucially, on this level of abstraction, the models are still amenable to analytical treatment using the framework of dynamical mean-field theory.The main focus of this thesis is to leverage the analytical tractability of random networks of point neurons in order to relate the network structure, and the neuron parameters, to the dynamics of the neurons—in physics parlance, to bridge across the scales from neurons to networks.More concretely, four different models are investigated: 1) fully connected feedforward networks and vanilla recurrent networks of rate neurons; 2) block-structured networks of rate neurons in continuous time; 3) block-structured networks of spiking neurons; and 4) a multi-scale, data-based network of spiking neurons. We consider the first class of models in the light of Bayesian supervised learning and compute their kernel in the infinite-size limit. In the second class of models, we connect dynamical mean-field theory with large-deviation theory, calculate beyond mean-field fluctuations, and perform parameter inference. For the third class of models, we develop a theory for the autocorrelation time of the neurons. Lastly, we consolidate data across multiple modalities into a layer- and population-resolved model of human cortex and compare its activity with cortical recordings.In two detours from the investigation of these four network models, we examine the distribution of neuron densities in cerebral cortex and present a software toolbox for mean-field analyses of spiking networks.
; 
|
The record appears in these collections: |
PDF
Fulltext