The authors address a generalized log-determinant semidefinite programming (SDP) optimization problem that covers various structures in Gaussian graphical models. The problem involves minimizing an objective function consisting of a linear term, a log-determinant term, and multiple ℓp-norm regularization terms, subject to linear equality constraints and a positive semidefinite constraint.
To solve this generalized problem efficiently, the authors propose a dual spectral projected gradient (DSPG) method. The key aspects of the method are:
The authors conduct numerical experiments on various log-likelihood minimization problems, including block constraints and multi-task structures. The results demonstrate the efficiency of the proposed method compared to existing approaches.
A otro idioma
del contenido fuente
arxiv.org
Ideas clave extraídas de
by Charles Namc... a las arxiv.org 10-01-2024
https://arxiv.org/pdf/2409.19743.pdfConsultas más profundas