Back to search

National Technical University of Athens (NTUA)

Ανάπτυξη παραλληλοποιήσιμων υπολογιστικών μεθοδολογιών χρονο-βηματισμού για την ανάλυση και τον σχεδιασμό δικτύων νευρώνων πολλαπλών χωροχρονικών κλιμάκων και πολύ μεγάλης κλίμακας

Abstract

dc:description

The brain with roughly 100 billion connected neurons constitutes indisputably, the most complicated and bigger scale network that we know. The operation of brain mainly, is based on the physiology and the interaction of neurons that determines the emergent dynamics which is developed in a macroscopic/clinically level. The time and space scale of behavior and the development of phenomena are extended by the operation of each one of neuron that includes important degree of heterogeneity (e.g. differents time scales of activation of ions in the membrane of neuron), in the interaction and transmission of signal (that it includes different space scales of transmission), in the network/body, up to the event of symptoms of illness in the individual. Due to the complexity of phenomena, the modem tendency in the mathematic description and the calculating simulation of dynamics of networks of neurons and more specifically, the dynamics of neurological dysfunctions is the growth of detail-microscopic models. Basic aim of this thesis is the growth of systematic calculating methodologies for bridging of the gap in time and space, between the level in which the operation of neurons can be available in detail description (microscopic-atomistic) and their interactions and the level in which there is desirable the analysis of presented dynamics and the planning of systems of control of symptoms in the organism (macroscopic level). In the frames of the present thesis, it was developed a calculating frame which allows in big scale simulators of dynamics of neurons of microscopic-atomistic level to immediately execute analysis in macroscopic level, without being essential the passage from an intermediary or a macroscopic level (conventional) description of the examined dynamics, via macroscopic equations in closed form (e.g. under the form of certain differential equations and/or differential-absolute equations). In this way, the proposed frame is general as for his application, adapting it suitably in each individualistic simulator. In the first part of the thesis there are being presented the main models of description of neurons, as well as the qualitative attributes that they present. Accent is given in the connection of dynamics with the way bum of neurons. Particularly, we examine thoroughly the more often presented dynamic behaviors of the neurons of Morris-Lecar type. Afterwards, developing the heart of theory of complexity, that is to say the theory of networks, there are being presented the most known forms of networks, Erdos and Renyi, Watts and Strogatz (WS) small world and finally the free scale networks Barabasi. Particular accent was given in their characteristics sizes as the characteristic length and the clustering coefficient. In capital 5 is developed a microscopic model combining the theory of networks and the biophysical description of neurons Morris-Lecar. The prototype part of this description is the appointment of the effect of the size of network in the macroscopic behavior of such systems: concretely, depending on the density of the network connections there are being observed different qualitatively situations with complicated non-linear dynamics that emerge with small changes in the density of connections. The capital 6 is focused on the main methodology of Equation-Free that is used for the systematic study and analysis of macroscopic dynamics from microscopic simulators which are used as “black boxes”. Also, it is presented the significance of the timestepper as well as the way in which this can be interlinked with advanced techniques of numerical analysis of tables-matrix free that are suitable for big scale systems, for the finding of points of balance, periodical solutions, the calculation of eigenvalues, the development of bifurcation diagram and finally, the study of rare events (by calculating the first time passage). Very important and original element in the analysis of problems of dynamics of neurons in networks is the connection of Equation Free methodology with the simulating annealing, a statistical technique of finding the most optimal solution, from the region of statistical physics. The main result of this hybrid method which is presented in capital 7, is that the dynamics of system, with use of microscopic simulator, it is led in a coarse-grained slow manifold of the “few” macroscopic variables. The new methodology that is being proposed is general under the significance that it can be applied in very big crowd of networks of neurons. In the two last capitals we set forth the main applications of the methodology that was developed in the capital 6 and 7. In the first application (capital 8) it is developed a cellular automata model on networks with constant connectivity and individualistic rules of interaction and transmission of signal. Leading the dynamics of microscopic simulator with the proposed methodology we calculate the steady states of the coarses veriables, their stability and the coarse grained bifurcation diagram. Finally, again using the hybrid method and constructing an Langevin equation with the Equation Free method, we find successfully the first mean time passage to escape of metastable situation. Finally, in the capital 9 we generalise the method in systems that are being develop in heterogeneous random networks with different per neuron degrees of interconnection. With the use of the proposed methodology it is analyzed the behavior of the system and are manufactured the coarse grained bifurcation diagrams on various prices of parameters as are the probability of activation of neurons and the connection probability of two neurons in the network. In the last capital it is being described the way in which has been parallelized the proposed methodology for the confrontation of increased calculating requirements, due to the big scale of problems that are being examined.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2009

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Spiliotis, Konstantinos
  • Σπηλιώτης, Κωνσταντίνος

Subjects

dc:subject × 10

Rights

Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/18276

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Spiliotis, Konstantinos; Σπηλιώτης, Κωνσταντίνος. Ανάπτυξη παραλληλοποιήσιμων υπολογιστικών μεθοδολογιών χρονο-βηματισμού για την ανάλυση και τον σχεδιασμό δικτύων νευρώνων πολλαπλών χωροχρονικών κλιμάκων και πολύ μεγάλης κλίμακας. National Technical University of Athens (NTUA), 2009. http://hdl.handle.net/10442/hedi/18276