SUPR
Robustness and Generalisation in Statistical Learning
Dnr:

NAISS 2024/22-1386

Type:

NAISS Small Compute

Principal Investigator:

Aleksandr Karakulev

Affiliation:

Uppsala universitet

Start Date:

2024-10-25

End Date:

2025-11-01

Primary Classification:

10201: Computer Sciences

Webpage:

Allocation

Abstract

This project focuses on the two important aspects of all modern statistical learning applications: robustness and generalisation. We are going to study how today's over-parameterized models (neural networks) can be trained effectively depending on the stability and convergence rates of the optimization procedures, model architecture, and objective formulation. With special focus, we are going to develop and test different probabilistic approaches to machine learning.