|
Abstract:
|
It is a fruitful and common practice to study the network structure constructed from a correlation matrix computed from fMRI time series data. With the advent of large public data sets with thousands sessions, it has become possible to ask meta-analytic questions about the variability of that network structure across sessions. There are challenging methodological issues that limit our ability to study the variation of network structure: registration, diversity of processing and analysis pipelines, and noise. In this presentation we will describe how the recently developed theory of dense graph limits provides a suitable representation space to study variability of network structures. We will present our findings based on analysis of the Human Connectome Project and MyConnectome datasets. We will also present our work how such network structure variation predicts neuroscientific correlates of interest. Time permitting we will also describe how we overcame some of the methodological challenges arising from working with data representations with no linear structure and how statistics with no linear structure lead to interesting theoretical work.
|
