Manifold Learning and Dimensionality Estimation for the Human Functional Connectome

The exploration of time-varying aspects of the human functional connectome (FC) is challenging because the high dimensionality of connectivity matrices precludes direct visual inspection in a meaningful manner. Dimensionality reduction helps circumvent this problem, yet effective application requires a-priori knowledge of the intrinsic dimension of the data and careful selection of algorithmic hyperparameters. Here, we first estimate the intrinsic dimension of FC data. Next, we use data with known cognitive state changes to evaluate the effectiveness of Laplacian Eigenmaps, T-SNE and UMAP to generate informative low dimensional representations of time-varying FC data for explorative and predictive purposes.