<ul><li>Graph learning, a key problem in graph signal processing, is crucial for applying the Fourier transform to non-Euclidean domains with unknown structures.</li><li>There is a growing interest in product graphs for modeling dependencies across different data dimensions, but current methods are limited in modeling diverse dependency structures.</li><li>This paper examines learning a Kronecker-structured product graph from smooth signals, which is more intricate than the Cartesian product but poses challenges in optimization.</li><li>An alternating optimization scheme is proposed to address the non-convex graph learning problem, with theoretical guarantees for convergence and superior performance shown in experiments.</li></ul>

Learning Kronecker-Structured Graphs from Smooth Signals

Discover more