Are you an EPFL student looking for a semester project?
Work with us on data science and visualisation projects, and deploy your project as an app on top of Graph Search.
We propose a physics-informed message-passing graph neural network (GNN) for learning the dynamics of springmass systems. The proposed method embeds the underlying physics directly into the message-passing scheme of the GNN. We compare the new scheme with conventional message passing and demonstrate the generalization capability of the method. Additionally, we infer the learned parameters of the edges and show that these parameters serve as explainable metrics for the learned physics. The numerical results indicate that the proposed method accurately learns the physics of the spring-mass systems.
Marco Picasso, Alexandre Caboussat, Maude Girardin
Demetri Psaltis, Carlo Gigli, Ahmed Ayoub
Alfio Quarteroni, Francesco Regazzoni, Stefano Pagani