Functional Role of Critical Dynamics in Flexible Visual Information Processing

http://nbn-resolving.de/urn:nbn:de:gbv:46-00107586-16
https://elib.suub.uni-bremen.de/peid=D00107586
https://elib.suub.uni-bremen.de/edocs/00107586-1.pdf
urn:nbn:de:gbv:46-00107586-16
Tömen, Nergis
2019
Universität Bremen: Physik/Elektrotechnik
Dissertation
Critical dynamics, phase transitions, visual system, neuronal network model, computational neuroscience, selective visual attention, feature integration, contour integration
Recent experimental and theoretical work has established the hypothesis that cortical neurons operate close to a critical state which signifies a phase transition from chaotic to ordered dynamics. Critical dynamics are suggested to optimize several aspects of neuronal information processing. However, although signatures of critical dynamics have been demonstrated in recordings of spontaneously active cortical neurons, little is known about how these dynamics are affected by task-dependent changes in neuronal activity when the cortex is engaged in stimulus processing. In fact, some in vivo investigations of the awake and active cortex report either an absence of signatures of criticality or relatively weak ones. In addition, the functional role of criticality in optimizing computation is often reported in abstract theoretical studies, adopting minimalistic models with homogeneous topology and slowly-driven networks. Consequently, there is a lack of concrete links between information theoretical benefits of the critical state and neuronal networks performing a behaviourally relevant task. In this thesis we explore such concrete links by focusing on the visual system, which needs to meet major computational challenges on a daily basis. Among others, the visual system is responsible for the rapid integration of relevant information from a large number of single channels, and in a flexible manner depending on the behavioral and environmental contexts. We postulate that critical neuronal dynamics in the form of cascades of activity spanning large populations of neurons may support such quick and complex computations. Specifically, we consider two notable examples of well-known phenomena in visual information processing: First the enhancement of object discriminability under selective attention, and second, a feature integration and figure-ground segregation scenario. In the first example, we model the top-down modulation of the activity of visuocortical neurons in order to selectively improve the processing of an attended region in a visual scene. In the second example, we model how neuronal activity may be modulated in a bottom-up fashion by the properties of the visual stimulus itself, which makes it possible to perceive different shapes and objects. We find in both scenarios that the task performance may be improved by employing critical networks. In addition, we suggest that the specific task- or stimulus-dependent modulations of information processing may be optimally supported by the tuning of relevant local neuronal networks towards or away from the critical point. Thus, the relevance of this dissertation is summarized by the following points: We formally extend the existing models of criticality to inhomogeneous systems subject to a strong external drive. We present concrete functional benefits for networks operating near the critical point in well-known experimental paradigms. Importantly, we find emergent critical dynamics only in the parts of the network which are processing the behaviourally relevant information. We suggest that the implied locality of critical dynamics in space and time may help explain why some studies report no signatures of criticality in the active cortex.
DDC
530
2019.08.14/13:11:36
Functional Role of Critical Dynamics in Flexible Visual Information Processing
[RefWorks] [Bibtex ]
[EndNote ] [RefMan(RIS) ]