Randomized Mixed Precision Algorithms for Large Scale Linear Algebra Problems with Laura Grigori

This talk discusses randomized mixed-precision algorithms introduced in recent years for solving large-scale linear and multilinear algebra problems. These methods rely on a powerful dimensionality reduction technique that enables the use of optimized kernels and mixed precision while providing numerical guarantees with high probability. Laura Grigori first discusses the sketching operation and present communication-optimal algorithms for different types of random sketching matrices, including dense, sparse, and structured matrices, then focuses on solving linear systems of equations and eigenvalue problems, as well as their integration into applications as materials science. These algorithms are becoming available in RandMixPack, a library developed for GPU and CPU architectures. Professor Laura Grigori holds the chair of high-performance numerical algorithms and simulation at EPFL and is the Head of the Laboratory for Simulation and Modelling at Paul Scherrer Institute, Switzerland. Her field of expertise is numerical linear, multilinear algebra, high-performance scientific computing and data analysis for challenging applications ranging from astrophysics to molecular simulations. She is a SIAM Fellow and the recipient of an ERC Synergy Grant. In 2024, she was awarded the SIAM Activity Group on Supercomputing Career prize in acknowledgement of her outstanding contributions to scientific computing, particularly communication-avoiding algorithms. For her work on communication-avoiding algorithms, she was awarded with her co-authors the 2016 SIAM Activity Group on Supercomputing Best Paper Prize. Watch previous SIAM Supercomputing Spotlight webinars at    • Supercomputing Spotlight   Learn more about SIAM Activity Group on Supercomputing at https://www.siam.org/get-involved/con... #SIAM #Supercomputing #MixedPrecision #Algorithms #LinearAlgebra #RandomSketching #Matrices #Orthogonalization #Eigenvalue #Orthogonalization