ARC@ORU research seminar series

The research seminar series ARC@ORU aims to raise awareness of the breadth of perspectives on AI, robotics and cybersecurity found at Örebro University, and to facilitate and inspire new collaborations.

SDE Matching: Scalable Variational Inference for Stochastic Differential Equations

September 12th, 13:15 - 14:30, Visual Lab, ARC (former "Labbet").

Registration: Please register here.

Speaker

Christian A. Naesseth, Assistant Professor of Machine Learning at the University of Amsterdam. Also a member of the Amsterdam Machine Learning Lab, the lab manager of the UvA-Bosch Delta Lab 2, and an ELLIS member.

Christans research interests span generative modeling, uncertainty quantification, reasoning, and machine learning, as well as their application to the sciences. He is currently working on generative models (diffusions, flows, generative flow networks, …), approximate inference (variational and Monte Carlo methods), probabilistic modelling (natural sciences, computer vision, health, etc.), uncertainty quantification and hypothesis testing (E-values, conformal prediction, …)

Previously, he was a postdoctoral research scientist with David Blei at the Data Science Institute, Columbia University. He completed his PhD in Electrical Engineering at Linköping University, advised by Fredrik Lindsten and Thomas Schön.

About the talk

The Latent Stochastic Differential Equation (SDE) is a powerful tool for time series and sequence modeling. However, training Latent SDEs typically relies on adjoint sensitivity methods, which depend on simulation, discretisation, and backpropagation through approximate SDE solutions, which limit scalability. In this work, we propose SDE Matching, a new simulation- and discretisation-free method for training Latent SDEs. Inspired by modern Score- and Flow Matching algorithms for learning generative dynamics, we extend these ideas to the domain of stochastic dynamics for time series and sequence modeling, eliminating the need for costly numerical simulations. The results demonstrate that SDE Matching achieves performance comparable to adjoint sensitivity methods while drastically reducing computational complexity.