<ul data-eligibleForWebStory="false"><li>This paper discusses learning undirected graphs from data collected at nodes within the graph signal processing framework.</li><li>The graph's topology is connected to the support of the conditional correlation matrix of the data.</li><li>The graph learning problem typically grows quadratically with the number of variables, posing challenges in high dimensions.</li><li>To address this, a graph learning framework utilizing a low-rank factorization of the conditional correlation matrix is proposed.</li><li>Tools necessary for applying Riemannian optimization techniques for this structure are derived to solve the optimization problems.</li><li>The proposal focuses on a low-rank constrained version of the GLasso algorithm for estimating a Gaussian graphical model using penalized maximum likelihood.</li><li>Experiments conducted on synthetic and real data show that this approach can achieve an efficient balance between dimensionality and performance.</li></ul>

Leveraging Low-rank Factorizations of Conditional Correlation Matrices in Graph Learning

Discover more