Yuji Nakatsukasa, "Randomised algorithms in numerical linear algebra and the column subset selection problem"

Thursday 20 Feb 2025, 14:30 → 15:30 Europe/Rome

Ex-ISEF/Building-Main Lecture Hall (GSSI)

Abstract:
Randomisation is among the most exciting developments in numerical linear algebra, and has led to algorithms that are fast, scalable, robust, and reliable.
The column subset selection problem (CSSP) is simple to state but not easy to solve, and has far-reaching consequences in a number of applications in computational mathematics. Several randomised and deterministic algorithms are now available for the CSSP.
In this talk we will review a few prominent topics in randomised NLA (low-rank approximation, least-squares problems, norm/trace estimation). We will then explore the CSSP from computational and theoretical viewpoints, highlighting their power in practical applications.