Title: Variable Clustering Using Distributionally Robust Nodewise Regression

Abstract: This research investigates a multi-factor block model designed for variable clustering, establishing a link between this approach and regularized subspace clustering by employing a distributionally robust variant of nodewise regression. To address the resulting optimization challenge, we introduce a convex relaxation technique, propose a data-driven methodology for determining the robust region's size, and design an Alternating Direction Method of Multipliers (ADMM) algorithm to facilitate efficient computation. The effectiveness of our proposed method is confirmed through comprehensive numerical experiments, which highlight its superior performance.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC