The Relationship between Multiscale Relevance and Network Properties in Simple Neural Networks of the Cortex

The lack of knowledge about the model underlying a certain generative process, as well as of the time scale on which the generated events unfold, requires us to think of a way to evaluate the informative content of a given sample without any assumptions on those missing information. Multiscale Relevance (MSR), drawing its theoretical formulation from the fields of statistical mechanics and information theory, aims at providing such an evaluation with a unique focus on the statistics of the sample. Its applications can be varied and across subjects, but up to the present moment, it has only been studied in reference to the spike trains generated by a population of neurons in the mEC area of a rat freely navigating a controlled environment (Cubero R.J., Marsili M., and Roudi Y., 2020). In this work, it has been established that the proposed metric, i.e. MSR, significantly correlates with other previously defined metrics that quantify the selectivity of neuronal response, as well as with other metrics generally applied (but not limited) to bursty signals. However, studying the properties of MSR in networks of biological neurons has, at the moment, some significant technological and ethical limitations, and consequently we find us in need for the generation of synthetic data from as many network models as possible. The present thesis aims at covering this gap, through the simulation of a simple model of the cortex, generating chaotic activity, across many different interaction conditions for its units, and through the study of how the MSR computed on the generated signal correlates with other metrics of the same, and with static graph measures (not depending on external input conditions) computed on the network.