Improving Matrix Exponential for Generative AI Flows: A Taylor-Based Approach Beyond Paterson--Stockmeyer

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Sastre, Jorge, Faronbi, Daniel, Alonso, José Miguel, Traver, Peter, Ibáñez, Javier, Lloret, Nuria
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918279606435840
author Sastre, Jorge
Faronbi, Daniel
Alonso, José Miguel
Traver, Peter
Ibáñez, Javier
Lloret, Nuria
author_facet Sastre, Jorge
Faronbi, Daniel
Alonso, José Miguel
Traver, Peter
Ibáñez, Javier
Lloret, Nuria
contents The matrix exponential is a fundamental operator in scientific computing and system simulation, with applications ranging from control theory and quantum mechanics to modern generative machine learning. While Padé approximants combined with scaling and squaring have long served as the standard, recent Taylor-based methods, which utilize polynomial evaluation schemes that surpass the classical Paterson--Stockmeyer technique, offer superior accuracy and reduced computational complexity. This paper presents an optimized Taylor-based algorithm for the matrix exponential, specifically designed for the high-throughput requirements of generative AI flows. We provide a rigorous error analysis and develop a dynamic selection strategy for the Taylor order and scaling factor to minimize computational effort under a prescribed error tolerance. Extensive numerical experiments demonstrate that our approach provides significant acceleration and maintains high numerical stability compared to existing state-of-the-art implementations. These results establish the proposed method as a highly efficient tool for large-scale generative modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2512_20777
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improving Matrix Exponential for Generative AI Flows: A Taylor-Based Approach Beyond Paterson--Stockmeyer
Sastre, Jorge
Faronbi, Daniel
Alonso, José Miguel
Traver, Peter
Ibáñez, Javier
Lloret, Nuria
Machine Learning
Numerical Analysis
65F60, 41A10, 65D15, 41-04, 68T07, 65Y20
The matrix exponential is a fundamental operator in scientific computing and system simulation, with applications ranging from control theory and quantum mechanics to modern generative machine learning. While Padé approximants combined with scaling and squaring have long served as the standard, recent Taylor-based methods, which utilize polynomial evaluation schemes that surpass the classical Paterson--Stockmeyer technique, offer superior accuracy and reduced computational complexity. This paper presents an optimized Taylor-based algorithm for the matrix exponential, specifically designed for the high-throughput requirements of generative AI flows. We provide a rigorous error analysis and develop a dynamic selection strategy for the Taylor order and scaling factor to minimize computational effort under a prescribed error tolerance. Extensive numerical experiments demonstrate that our approach provides significant acceleration and maintains high numerical stability compared to existing state-of-the-art implementations. These results establish the proposed method as a highly efficient tool for large-scale generative modeling.
title Improving Matrix Exponential for Generative AI Flows: A Taylor-Based Approach Beyond Paterson--Stockmeyer
topic Machine Learning
Numerical Analysis
65F60, 41A10, 65D15, 41-04, 68T07, 65Y20
url https://arxiv.org/abs/2512.20777