Researcher Wanli Qiao redefines density ridge extraction in ML

Machine learning researchers have long relied on the Subspace Constrained Mean Shift (SCMS) algorithm to extract density ridges—low-dimensional representations of high-dimensional data. For over a decade, the prevailing assumption was that SCMS trajectories converge to a 'static ridge,' defined by the density gradient and Hessian eigenvectors. In a groundbreaking 40-page paper submitted to arXiv on August 5, 2026, statistician Wanli Qiao dismantles this assumption, revealing that the static definition fails to account for eigenspace rotation during the algorithm's vector field flow.

Qiao introduces the 'stable ridge'—a novel geometric structure grounded in dynamical systems theory and the Jacobian of the projected density gradient—as the true theoretical target of SCMS. His generalized framework, using constant step sizes, establishes uniform R-linear convergence and topological surjectivity onto the stable ridge. He also derives convergence rates using Hausdorff distance and demonstrates that the original SCMS suffers from polynomial-time complexity due to implicit coupling between step size and smoothing bandwidth. The proposed solution offers statistically consistent and computationally efficient density ridge extraction, reshaping how high-dimensional data is analyzed in machine learning applications.