Inference for Newton Methods with Accelerated Sketch-and-Project via Random Scaling

DOI10.48550/arXiv.2609.12421MaRDI QIDQ7345518arXivFDO

Authors Elizaveta Rebrova, Michał Dereziński, Sen Na, Xinchen Du

Publication date 11 September 2026

Published in arXiv

Tldr This paper proposes an accelerated sketch-and-project solver and analyzes its theoretical properties for Newton methods in large-scale optimization problems. The main contribution is an asymptotic normality result that enables online inference via random scaling, bypassing explicit covariance estimation and allowing for asymptotically valid tests even when the number of observations is comparable to or smaller than the dimensionality.

Full work available at URL https://doi.org/10.48550/arXiv.2609.12421

arXiv classification

cs.LGmath.NAmath.OCmath.STstat.ML

This page was built for publication: Inference for Newton Methods with Accelerated Sketch-and-Project via Random Scaling

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q7345518)