Dynamic Graph Learning: A Structure-Driven Approach

The purpose of this paper is to infer a dynamic graph as global (collective) model time-varying measurements at set network nodes. This captures both pairwise well higher order interactions (i.e., more than two nodes) among the motivation work lies in search for connectome which properly brain functionality across all regions brain, and possibly individual neurons. We formulate it an optimization problem, quadratic objective functional tensor information observed node signals over short time intervals. proper regularization constraints reflect smoothness other dynamics involving underlying graph’s Laplacian, evolution graph. resulting joint solved by continuous relaxation weight parameters introduced novel gradient-projection scheme. While may be applicable any time-evolving data (e.g., fMRI), we apply our algorithm real-world dataset comprising recorded activities cells. shown not only viable but also efficiently computable.