Multifidelity Graph Neural Networks for Efficient and Accurate Mesh-based PDE Surrogate Modeling

This paper introduces two multifidelity graph neural network strategies for mesh-based PDE surrogate modeling, including a hierarchical coarse-to-fine architecture and a curriculum-learning-based training scheme. The work shows how low-fidelity and high-fidelity simulations can be combined to reduce training cost while improving robustness, generalization, and accuracy on both 2D mechanics problems and large-scale 3D CFD examples. The paper was also selected for the journal cover, as noted in the updated CV.

Recommended citation: Taghizadeh, M., Nabian, M. A., & Alemazkoor, N. (2024). "Multifidelity Graph Neural Networks for Efficient and Accurate Mesh-based Partial Differential Equations Surrogate Modeling." Computer-Aided Civil and Infrastructure Engineering.