Scalable numerical linear algebra for science and machine learningLecture

by Dr David Persson

Friday Nov 6, 2026, 10:15 AM → 11:00 AM Europe/Berlin

Endenicher Allee 60/1-016 - Lipschitzsaal (Mathezentrum)

Large-scale computational methods have become central to modern science. Methods from statistics, machine learning, and physical simulation have transformed fields from weather prediction to astrophysics to quantum mechanics. Despite the diversity of computational methods used in such fields, a surprising number of our most important tools are based on a single shared computational core: numerical linear algebra (NLA). Research in NLA impacts a huge range of downstream computational methods by improving algorithms for basic matrix computations like linear system solving, matrix multiplication, and data compression. Among these, matrix compression has become especially important for large-scale problems, where storing or computing with a matrix directly may be prohibitively expensive. This talk focuses on replacing such matrices by structured approximations that are cheaper to store and compute with. Randomization provides a powerful approach for computing these approximations, even when a matrix is accessible only through matrix-vector products. We will discuss recent results on hierarchical matrix compression, together with broader directions in NLA for scientific computing and machine learning.