The Polar Express: Optimal Matrix Sign Methods and their Application to the Muon Algorithm

ICLROral2026

Authors
Noah Amsel, David Persson, Christopher Musco, Robert M. Gower
Affiliation
NYU, New York University
Venue
ICLR 2026
Track
Oral

TL;DR

We introduce a GPU-friendly algorithm for computing the polar decomposition of a matrix to low accuracy that is optimal in its class. This improves Muon.

Opening excerpt from the authors’ abstract. source

Read the paper

← All ICLR 2026 Oral papers · Browse the whole archive