MeshStudioAI

MeshStudioAI is a collection of AI tools for adaptive meshing in the context of finite element methods (FEM). It builds on the official implementation of G-Adaptivity: a GNN-based approach to adaptive mesh relocation for FEM.

The full repository is available on GitHub.

Functionality

With this package you can:

• Use the pretrained G-Adaptivity framework on your own FEM problem
• Train a mesh relocation model (typically MeshAdaptor)
• Evaluate trained models on precomputed datasets
• Plot qualitative and quantitative results
• Log experiments to Weights & Biases (W&B)

Quickstart

To run your first mesh adaptivity example please see the Installation guide, and the First demo notebook.

License and citation

This open-source version of our code is licensed under Apache 2.0. If you use this work, please cite:

@inproceedings{Rowbottom_G-Adaptivity_optimised_graph-based_2025,
    author = {Rowbottom, James and Maierhofer, Georg and Deveney, Teo and Müller, Eike Hermann and Paganini, Alberto and Schratz, Katharina and Lio, Pietro and Schönlieb, Carola-Bibiane and Budd, Chris},
    booktitle = {Proceedings of the Forty-second International Conference on Machine Learning},
    title = {{G-Adaptivity: optimised graph-based mesh relocation for finite element methods}},
    year = {2025}
}