An Augmented Backward-Corrected Projector Splitting Integrator for Dynamical Low-Rank Training
SIAM Journal on Mathematics of Data Science · 30.09.2026 · 8 (3) · 820-849
DOI: 10.1137/25m1730673 ↗Kurz gefasst
Fachartikel in der Zeitschrift SIAM Journal on Mathematics of Data Science, online erschienen am 07.08.2026. Verfasst von Jonas Kusch und 2 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 59 Literaturangaben. Gefördert von: Helmholtz Information & Data Science School for Health, Helmholtz Information and Data Science Academy, U.S. Department of Energy.
Abstract (Originalsprache)
Layer factorization has emerged as a widely used technique for training memory-efficient neural networks. However, layer factorization methods face several challenges, particularly a lack of robustness during the training process. To overcome this limitation, dynamical low-rank training methods have been developed, utilizing robust time integration techniques for low-rank matrix differential equations. Although these approaches facilitate efficient training, they still depend on computationally intensive QR and singular value decompositions of matrices with small rank. In this work, we introduce a novel low-rank training method that reduces the number of required QR decompositions. Our approach integrates an augmentation step into a projector-splitting scheme, ensuring that a subsequence of the projected gradient converges to zero in probability. We provide a rigorous theoretical analysis of the proposed method and demonstrate its effectiveness across multiple benchmarks.
Förderung
- Helmholtz Information & Data Science School for Health
- Helmholtz Information and Data Science Academy
- U.S. Department of Energy
Zitieren
APA
Kusch, J., Schotthöfer, S., & Walter, A. (2026). An Augmented Backward-Corrected Projector Splitting Integrator for Dynamical Low-Rank Training. SIAM Journal on Mathematics of Data Science, 8(3), 820-849. https://doi.org/10.1137/25m1730673
Vancouver
Kusch J, Schotthöfer S, Walter A. An Augmented Backward-Corrected Projector Splitting Integrator for Dynamical Low-Rank Training. SIAM Journal on Mathematics of Data Science. 2026;8(3):820-849. doi:10.1137/25m1730673
BibTeX
@article{Kusch2026,
author = {Kusch, Jonas and Schotthöfer, Steffen and Walter, Alexandra},
title = {An Augmented Backward-Corrected Projector Splitting Integrator for Dynamical Low-Rank Training},
journal = {SIAM Journal on Mathematics of Data Science},
year = {2026},
volume = {8},
number = {3},
pages = {820--849},
doi = {10.1137/25m1730673}
}
EUVIMED