{"id":"arxiv/2610.02633","url":"https://arxiv.org/abs/2610.02633","tags":["math.ST","stat.ME"],"score":0,"title":"High-Dimensional Regularization of the Spatial Sign Covariance Matrix for Robust Shape Estimation","domain":"math.ST","source":"arxiv","authors":["Jonas Elmerraji","James C. Spall","Mateo Díaz"],"reviews":0,"subject":"math.ST","summary":"Covariance estimation is a key component of many applications in system identification and data-driven control. Although heavy-tailed distributions may lack a covariance matrix to estimate, the shape matrix provides a well-defined, scale-free generalization for the broad family of elliptical distributions. In this setting, practitioners often employ Tyler's M-estimator (TME), which is defined implicitly and is computed using a fixed-point iteration. The spatial sign covariance matrix estimator (SSCM) offers a much simpler alternative: it corresponds to one iteration of TME. Yet, the SSCM has…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-10-02","reactions":{}}