COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Parkina, Uliana, Rakhuba, Maxim
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914419521355776
author Parkina, Uliana
Rakhuba, Maxim
author_facet Parkina, Uliana
Rakhuba, Maxim
contents Recent studies suggest that context-aware low-rank approximation is a useful tool for compression and fine-tuning of modern large-scale neural networks. In this type of approximation, a norm is weighted by a matrix of input activations, significantly improving metrics over the unweighted case. Nevertheless, existing methods for neural networks suffer from numerical instabilities due to their reliance on classical formulas involving explicit Gram matrix computation and their subsequent inversion. We demonstrate that this can degrade the approximation quality or cause numerically singular matrices. To address these limitations, we propose a novel inversion-free regularized framework that is based entirely on stable decompositions and overcomes the numerical pitfalls of prior art. Our method can handle possible challenging scenarios: (1) when calibration matrices exceed GPU memory capacity, (2) when input activation matrices are nearly singular, and even (3) when insufficient data prevents unique approximation. For the latter, we prove that our solution converges to a desired approximation and derive explicit error bounds.
format Preprint
id arxiv_https___arxiv_org_abs_2507_07580
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation
Parkina, Uliana
Rakhuba, Maxim
Machine Learning
Computation and Language
Numerical Analysis
65F55, 68T50
Recent studies suggest that context-aware low-rank approximation is a useful tool for compression and fine-tuning of modern large-scale neural networks. In this type of approximation, a norm is weighted by a matrix of input activations, significantly improving metrics over the unweighted case. Nevertheless, existing methods for neural networks suffer from numerical instabilities due to their reliance on classical formulas involving explicit Gram matrix computation and their subsequent inversion. We demonstrate that this can degrade the approximation quality or cause numerically singular matrices. To address these limitations, we propose a novel inversion-free regularized framework that is based entirely on stable decompositions and overcomes the numerical pitfalls of prior art. Our method can handle possible challenging scenarios: (1) when calibration matrices exceed GPU memory capacity, (2) when input activation matrices are nearly singular, and even (3) when insufficient data prevents unique approximation. For the latter, we prove that our solution converges to a desired approximation and derive explicit error bounds.
title COALA: Numerically Stable and Efficient Framework for Context-Aware Low-Rank Approximation
topic Machine Learning
Computation and Language
Numerical Analysis
65F55, 68T50
url https://arxiv.org/abs/2507.07580