Implementing Graph Neural Networks in Modulus

Hello, I’d love to ask a bit open-ended question, but do you think that one can write a custom code to extend Modulus to support also graph neural networks? In particular, GNNs that can be used for simulating physics?

I was implementing a custom framework before stumbling on Modulus that followed a work done in this area from DeepMind (Learning to simulate). Afterwards we switched to Modulus since it has lot of stuff for solving practical problems in physics, but I just wonder if it’s possible to implement something like MeshGraphNets mentioned in DeepMind’s work in Modulus.

You’ll be happy to know that Graph Networks will be in Dec release of Modulus (22.12) and the DeepMind paper is one of the networks among some others that would be available.

Very interesting! Looking forward!