Analysing Neuronal Network Architectures : From Weight Distributions to Structure and Back

The connections cortical neurons forms are different in each individual human or animal. Although there are known and determined large scale connections between areas of the brain that are common across individuals, the local connectivity on smaller scales varies between individuals. Connections between neurons in a single cortical column are seemingly random and were thus modeled in theoretical studies using the principle of sparse random networks. In this thesis I investigate how the simplifications of sparse random networks affect the behaviour and plausibility of the network. First, I focus on the global weight distribution in random sparse networks and test the impact of various weight distributions on network excitability. Randomsparse networks also commonly assume an independent distribution of connections in the network. Experimental results indicate that this assumption is not true in biological neuronal networks. In the second part of this thesis it is shown how changes in degree distributions of the network can be employed to improve the similarity of random networks to biological observations. A third aspect studied in this work is the impact of connection strengths on a local level. Network responses to larger stimuli in in vitro experiments are not reproducible by classical sparse randomnetworks. It is shown how changes in the distributions of local connection strengths can be used to improve the network response behaviour with respect to these experimental findings. Finally, these network adjustments are combined and the adjusted networks are tested on further data of in vivo recordings. The adjusted networks show a biologically more plausible behaviour than the random sparse network topologies in all tested scenarios.