An overview of low-rank compression techniques for the approximation of dynamical problems
by
OHSA/E13
Abstract:
The dynamical low-rank approximation is a recent technique for solving large-scale problems with a low-memory footprint. It has been successfully applied to problems from quantum chemistry, plasma physics, radiation therapy, stochastic simulations, and recently to train models in machine learning.
My talk will start by introducing the basic mathematical background and show some of the existing applications. Then, I will present a comprehensive overview of the algorithms I have developed during my PhD and briefly mention the research directions I am following now. The applications of these new methods are still largely unexplored, so if you are looking for new original methods to speed up your simulations of time-dependent problems: you should come to my talk!
Bio:
Benjamin received his Bachelor’s degree and Master’s degree in Mathematics, and did his PhD in Applied Mathematics at the University of Geneva (Switzerland). His doctoral research focused on the development of new algorithms for the low-rank compression of large-scale problems. He mainly developed and analyzed a low-rank parallel-in-time algorithm and a set of low-rank methods for stiff problems. In March 2025, he joined Laura Grigori’s group (HPNALGS) at PSI as a postdoctoral fellow. He is currently studying the potential of randomization techniques for numerical linear algebra for developing even more robust algorithms. He is also looking forward to engaging in collaborations at PSI to find new applications for the low-rank compression techniques he has already developed.
The Laboratory for Simulation and Modeling