Title:	Graph-based, spatial, temporal, and generative machine learning
DNr:	Berzelius-2026-109
Project Type:	LiU Berzelius
Principal Investigator:	Fredrik Lindsten <fredrik.lindsten@liu.se>
Affiliation:	Linköpings universitet
Duration:	2026-04-01 – 2026-10-01
Classification:	10210
Keywords:

Abstract

This proposal represents a coordinated research effort of the research group lead by PI Professor Fredrik Lindsten, constituting five projects that involve five PhD students, one postdoctoral researcher, and an adjunct associate professor. Together, the group advances method development in graph‑based, spatio‑temporal, and generative machine learning, with applications spanning weather and climate modelling, materials science, and scientific generative modelling. The project is a continuation of previous Berzelius projects, which have resulted in publications at top machine learning venues including NeurIPS, ICLR, ICML, and AISTATS. The projects investigate method development related to probabilistic weather modelling, data assimilation, flow‑based and diffusion‑based generative modelling, interacting particle systems for controlled generation, machine‑learning‑driven inverse materials design, and geo‑spatial conditioning mechanisms for weather downscaling. Several efforts are carried out in collaboration with partners including SMHI, the Danish Meteorological Institute, the Finnish Center for Artificial Intelligence, and the California Institute of Technology. This research is supported by multiple sources, including the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation via the WASP NEST project Multi‑dimensional Alignment and Integration of Physical and Virtual Worlds, the WASP‑WISE NEST project 2DFound, and a WASP Academic PhD project. Additional financial support comes from the Swedish Research Council through the project Machine Learning for Complex Spatio-Temporal Processes (contract number 2024-05011)), as well as from the Excellence Center at Linköping–Lund in Information Technology (ELLIIT). Access to large‑scale GPU resources is essential for training and evaluating large scale models, enabling the group to push methodological boundaries and continue producing high‑impact research in AI and machine learning.