Answers are grounded in this paper and cite their sources.
Demonstrates improved estimation accuracy in matrix-variate data, suggesting enhanced statistical modeling techniques.
Matrix-variate data is prevalent in numerous scientific and engineering domains, including medical imaging and geographic information systems. It is widely acknowledged that matrix-variate data encompasses significant structural information that can enhance the performance of statistical models. Nuclear-norm regularization has emerged as a leading technique for capturing low-rank structures in various matrix-variate problems. However, solutions to nuclear-norm regularized models often depend on singular value decomposition of the data matrix, which is susceptible to data perturbations. We propose decomposing the matrix parameter into a low-rank-plus-sparse structure within matrix-variate regression models and introduce a doubly regularized regression (DRR) criterion, which bolsters the robustness of nuclear-norm regularized models. From a computational standpoint, we have developed a scalable algorithm to address the low-rank-plus-sparse problem, achieving optimal convergence rates among first-order methods. On the statistical front, we have established non-asymptotic theoretical guarantees for exact structure recovery and estimation error bounds under certain regularity conditions. The numerical performance of our proposed method has been validated through extensive simulations and an analysis of ElectroEncephaloGraphy (EEG) data. An R package drrglm available on CRAN is ready to facilitate the use of the proposed DRR method.
No takes yet. Share an insight, caveat, or question.
Xu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: