Weighted Low-rank Approximation via Stochastic Gradient Descent on Manifolds

Conglong Xu, Peiqi Yang, Hao Wu · 2025

Read Weighted Low-rank Approximation via Stochastic Gradient Descent on Manifolds by Conglong Xu, Peiqi Yang, Hao Wu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-14174.txt.

Description

Conglong Xu and co-authors solve a weighted low-rank approximation problem using gradient descent that moves on a curved manifold. Their 2025 paper deals with a core data task: approximating a large matrix by a simpler low-rank one, where some entries matter more than others. Doing the descent on the manifold of fixed-rank matrices respects the problem's natural geometry.

To make the stochastic method trustworthy, the authors first prove a convergence theorem for retraction-based descent on manifolds under confinement conditions. They then test the algorithm on data from the Netflix Prize training set. It beat a Euclidean-space version of the same approach. They also compare accelerated line search on the manifold against its flat-space counterpart, rounding out the case for geometric methods.

Catalog record

CREATORConglong Xu, Peiqi Yang, Hao Wu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE178 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-14174

Browse the full library · About the Tower of Babel