Truncated LSQR for matrix and tensor equations

08.07.2026, 12:00  –  Campus Golm, Building 9, Room 1.22
Forschungsseminar Numerische Analysis

Lorenzo Piccinini (University of Bologna)

We are interested in the numerical solution of multi-term matrix and tensor least squares problems. We focus on the large-scale setting, where the large dimensions lead to memory issues and higher computational complexity. We propose a truncated matrix- and tensor-oriented version of the well-known LSQR algorithm. In both cases, we present how to derive the algorithm and how truncation affects it. To illustrate their effectiveness, we present numerical experiments for dictionary learning and information retrieval.