Niclas Popp
I am an ELLIS Doctoral student at the University of Tübingen and the Bosch Center for Artificial Intelligence. The work on randomized diagonal estimation presented at JuliaCon 2024 was carried out in the course my master thesis at TU Munich and KTH Stockholm (supervised by Claudio Mayrink Verdun, Felix Krahmer and Elias Jarlebring).
Session
Implicit diagonal estimation is a long-standing problem that is concerned with approximating the diagonal of a matrix (function) that can only be accessed through matrix-vector products. We present RandomizedDiagonalEstimation.jl, which implements existing and novel randomized methods for accomplishing this task in a scalable way. The practical applications of the provided algorithms extend across diverse domains such as network science, material science, optimization and machine learning.