Collaborative Research: Elements: A Cyberlaboratory for Randomized Numerical Linear Algebra

University Of Wisconsin-Madison, Madison WI

Investigators

Abstract

Modern scientific and engineering computations require processing data at speeds and scales that overwhelm traditional linear algebra algorithms, which creates a substantial bottleneck in solving cutting-edge scientific and engineering problems. To overcome this challenge, linear algebra algorithms are being augmented with randomization to compress input data, a technique that has shown tremendous promise in theory. Unfortunately, these theoretical advances have not transformed into practical implementations because of a lack of a cyberinfrastructure that enables rapid and reproducible prototyping of such algorithms. To address this deficit, this project creates an innovative cyberinfrastructure to enable rapid implementation and reproducible comparison of these randomized linear algebra algorithms. Additionally, this project enables easily testing randomized linear algebra algorithms in scientific computations prior to expensive software development. Thus, this project is poised to support the solution of cutting-edge computational science and engineering problems. Moreover, this project utilizes the cyberinfrastructure to deliver publicly accessible educational modules on randomized linear algebra theory and implementation. This project develops a cyberinfrastructure for randomized algorithms with a focus on algorithms for linear algebra problems. This project maps our innovative theoretical modularization of randomized linear algebra algorithms to computational kernels, which can be readily composed and extended to rapidly develop novel, tailored methods for specific scientific computations. Moreover, this cyberinfrastructure includes a framework for reproducibly testing algorithms, and enables benchmarking codes prior to deployment. This cyberinfrastructure demonstrates its power by directly addressing cutting-edge data problems in climate applications. The project emphasizes best scientific software practices to develop a sustainable, community-engaging and impactful cyberinfrastructure. This award by the NSF Office of Advanced Cyberinfrastructure is jointly supported by the Division of Mathematical Sciences. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.