GPU-Enabled Large-Scale Optimization Using Randomized Linear Algebra — arXiv 中文翻译

GPU-Enabled Large-Scale Optimization Using Randomized Linear Algebra英文标题：GPU-Enabled Large-Scale Optimization Using Randomized Linear Algebra作者：Pratik Rathore, Zachary Frangella, Parth Nobel, Xuning Hu, Madeleine UdellarXiv ID：2609.08136 | 分类：cs.LG | 发表：2026-09-08许可：CC-BY中文摘要摘要本文介绍了rlaopt，一个基于PyTorch的软件包，用于使用随机数值线性代数（RandNLA）进行大规模优化和科学计算。{{NL}}尽管基于RandNLA的算法取得了实质性进展，但很少有实现能够将GPU加速与用于指定优化问题的简单接口结合起来。{{NL}}rlaopt通过为正定线性系统以及带约束和正则项的凸经验风险最小化提供GPU启用的求解器，填补了这一空白。{{NL}}这些求解器使用RandNLA来加速共轭梯度（NyströmPCG）、算子分裂（NysADMM）和随机梯度方法（SAPPHIRE）。{{NL}}此外，rlaopt包含一种建模语言，允许用户使用自然的数学语法来指定问题。英文摘要This paper introduces rlaopt, a PyTorch-based package for large-scale optimization and scientific computing using randomized numerical linear algebra (RandNLA). Despite substantial progress in RandNLA-based algorithms, few implementations combine GPU acceleration with a simple interface for specifying optimization problems. rlaopt addresses this gap by providing GPU-enabled solvers for positive-definite linear systems and convex empirical risk minimization with constraints and regularizers. These solvers use RandNLA to accelerate conjugate gradient (NystromPCG), operator splitting (NysADMM), and stochastic gradient methods (SAPPHIRE). Moreover, rlaopt includes a modeling language that lets users specify problems using natural mathematical syntax. rlaopt automatically checks compatibility with the selected solver and performs the required problem decomposition. The solvers also support differentiation through their iterations, enabling applications such as hyperparameter tuning. Experiments on ridge regression, bounded multinomial logistic regression, and bounded elastic net identify when randomized preconditioning improves performance and demonstrate substantial speedups from GPU execution. The package is open-source under an Apache license, with source code at https://github.com/udellgroup/rlaopt and version 0.1.0 available on PyPI.查看完整双语翻译 →

正在跳转到翻译阅读页… 如果没有自动跳转，请点击这里。