Title:	Learning non-linear multimodal asymptotically stable dynamical systems from data
DNr:	Berzelius-2025-218
Project Type:	LiU Berzelius
Principal Investigator:	Rodrigo Perez <rpd@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2025-06-26 – 2026-01-01
Classification:	10210
Keywords:

Abstract

This project focuses on developing novel machine learning techniques for learning asymptotically stable multimodal non-linear dynamical systems from data. The objective is to introduce inductive biases into deep neural networks modeled via flow matching, which are highly expressive models and have successfully represented multimodal data. Such inductive biases do not exist so far for systems modeled via flow matching, so it would be the first of its kind.