A1495
Title: Canonical mutual information under meta-elliptic symmetry
Authors: Sarbojit Roy - King Abdullah University of Science and Technology (Saudi Arabia) [presenting]
Abstract: Understanding brain network connectivity requires interpretable models that can capture complex neural dependencies. While canonical correlation analysis is restricted to linear associations, more general dependence measures such as mutual information (MI) can detect both linear and nonlinear relationships but often lack interpretability at the covariate level. We propose a semiparametric, interpretable framework to quantify the global association between two brain regions using MI. Our network-based estimator is consistent under meta-elliptic symmetry of the covariates and can highlight key drivers of connectivity. These interpretability approaches are especially suited for neuroscience applications, where identifying and explaining connectivity patterns is essential.
