Learning the Dynamics of Physical Systems from Sparse Observations with Finite Element Networks
Physical systems are at the core of many scientific and engineering efforts and the scientific field has explored many of them in great detail. Nonetheless, machine learning efforts to model these systems have been quite separated from the partial differential equations commonly used to describe their dynamics and the numerical methods that are used to solve these equations.
In this talk, I will describe a new model for spatio-temporal forecasting that we have introduced in joint work with Stephan Günnemann. The proposed graph neural network is closely related to the finite element method and can be equipped with intrinsic knowledge of partial differential equations. I will explain the derivation of the model and present empirical results from our paper.
Marten Lienen is a PhD student with Professor Stephan Günnemann in the Data Analytics and Machine Learning Group in the Department of Informatics at the Technical University of Munich.
In his research, Marten combines machine learning with numerical approaches from computational science to develop principled machine learning models for spatio-temporal forecasting. Before joining DAML, Marten Lienen worked in autonomous driving at BMW.
To become a member of the Rough Path Interest Group, register here for free.