Title:	Generative machine learning and amortized optimization
DNr:	Berzelius-2026-187
Project Type:	LiU Berzelius
Principal Investigator:	Jens Sjölund <jens.sjolund@it.uu.se>
Affiliation:	Uppsala universitet
Duration:	2026-06-26 – 2027-01-01
Classification:	10203
Homepage:	https://jsjol.github.io/
Keywords:

Abstract

This project requests a continuation of Berzelius-2025-399 with increased compute and storage allocations to support an integrated research program in numerical optimization, machine learning, and generative modeling. The group conducts methodological and applied research across three interrelated areas: (i) large-scale numerical optimization and learning-to-optimize methods, including GNN-accelerated solvers; (ii) diffusion and flow-based generative models with applications in computer vision and inverse problems; and (iii) Bayesian experimental design with applications in materials science. The requested resources support several PhD students and postdocs working on these topics. During the previous allocation period (November 2025 - June 2026), we substantially exceeded the allocated compute on both Ampere (188% of allocation) and Hopper (102%), while storage usage reached 97% of the allocated capacity (even after being granted a 50% increase mid-way through the project period). The increased allocation requested here reflects this demonstrated demand, which we expect to persist as ongoing research lines mature and new projects ramp up.