Great work!
Could you please provide some comparisons between the proposed GNN based solution and INR (implicit neural representation, sometimes called the coordinate network, neural fields or physics informed network), which is also a popular method for optimizing PDEs? e.g., the SIREN.
Thank you!