SDE Matching:Simulation-Free Learning of Stochastic Dynamics by Dr. Christian Andersson Naesseth

19 Jan 2026 10.00 AM - 11.00 AM ESR10 Current Students, Industry/Academic Partners, Prospective Students

Abstract
Stochastic differential equations (SDEs) provide a flexible framework for  modeling time series, dynamical systems, and sequential data. However,learning SDEs from data typically relies on adjoint sensitivity methods, which require repeated simulation, time discretization, and back propagation through approximate SDE solvers, leading to significant
computational overhead and limited scalability. We introduce SDE Matching, a simulation- and discretization-free
approach for learning stochastic dynamics directly from data. Building on recent advances in score matching and flow matching for generative modeling, we extend these ideas to the dynamical setting, enabling direct learning of SDE drift and diffusion terms without numerical simulation. SDE Matching replaces solver-based training with a regression-like
objective defined on transformed data samples, eliminating the need for backpropagation through stochastic trajectories. Empirically, SDE Matching achieves accuracy comparable to adjointbased methods while substantially reducing computational cost, offering a scalable alternative for learning stochastic dynamical systems.

 

Biography
Christian A. Naesseth is an Assistant Professor of Machine Learning at the University of Amsterdam and a member of the Amsterdam Machine Learning Lab (AMLab). He is also a member of ELLIS and serves as lab manager for the UvA–Bosch Delta Lab 2. His research lies at the intersection of generative modeling, stochastic dynamics, and probabilistic inference. He works on methods such as diffusion models, flow-based and simulation-free approaches to learning dynamical systems, and approximate inference using variational and Monte Carlo techniques. He is particularly interested in developing scalable probabilistic methods with applications in the natural sciences, computer vision, and  healthcare. Prior to joining the University of Amsterdam, Christian was a postdoctoral research scientist with David Blei at Columbia University’s Data Science Institute. He received his PhD in Electrical Engineering from Linköping University, where he was advised by Fredrik
Lindsten and Thomas Schön.

Categories