Sparse structure learning for consensus network systems via sum-of-absolute-values regularization
Fumiya Matsuzaki, Rio Yotsumoto, Takuya Ikeda · 2023
This paper proposes an identification method of network topology of consensus dynamical systems based on sparse structure learning. Assuming that the topology is binary weighted and undirected, we formulate the topology identification problem as a sparse maximum likelihood estimation problem, by employing the idea of sum-of-absolute-values optimization. For the computation, we apply a proximal gradient method and provide its closed form. The effectiveness is illustrated with a numerical example.