Some BIGMATH challenges
BIGMATH is a current European Industrial Doctorate project carried out by colleagues and companies in Milan, Lisbon, Novi Sad, and Eindhoven.
In this talk, we will review some current work in this project, mostly in two directions. The first challenge is to find quality low-rank interpolative or CUR decompositions for a given data matrix. These approximations currently attract attention since they provide an easy interpretation in terms of data points or variables.
The second challenge is the stepsize selection for gradient methods to solve large-scale numerical optimization problems. We will review topics such as secant condition, Barzilai-Borwein and harmonic stepsizes, and limited memory methods.
We will also highlight some open problems. This is joint work with Giulia Ferrandi and Perfect Gidisu.