Applications of Network Representation Learning to Identify Associations of Brain Function with Political Ideology
Conference
64th ISI World Statistics Congress
Format: IPS Abstract
Keywords: graph representation learning;, network, network-modelling
Session: IPS 315 - Recent advances in modeling and analysis of large high-dimensional networks
Monday 17 July 2 p.m. - 3:40 p.m. (Canada/Eastern)
Abstract
Emerging research examining functional connectivity (i.e, synchrony or correlation of activity between multiple brain regions) has begun investigating the neural underpinnings that drive political ideology, political attitudes, and political actions. This session will explore network-driven large sample, whole-brain analyses of functional connectivity across common fMRI tasks to identify associations with political ideology in a large sample from The Ohio State University Wellbeing project. I will discuss how to use network representation learning (NRL) and other classic data science techniques (graph-theoretic convolutional neural networks + principal component analysis + penalized regression techniques) to explore functional connectivity associations with political ideology. With NRL, I find that functional connectivity data can be used to accurately differentiate conservatives from liberals and that functional connectivity augments traditional survey-based predictions of political ideology. This case study motivates the use of NRL for downstream learning tasks like regression and classification.