GAOT
A geometry-aware operator transformer for PDEs on arbitrary domains
Paper title Geometry Aware Operator Transformer as an Efficient and Accurate Neural Surrogate for PDEs on Arbitrary Domains
Paper arXiv 2505.18781PDF
A multiscale attentional graph neural operator encodes inputs on any point cloud or mesh onto tokens of a coarse latent grid, together with geometry embeddings that summarise the local point distribution; a vision transformer processes the patched tokens, and a matching graph decoder returns the field at any query point, so a model trained on under 10% of a car's surface points can still be queried on the full surface, at some loss of accuracy.
- Published
- NeurIPS
- Data
- DrivAerNet++, DrivAerML
Share on LinkedIn, X or Bluesky, or or .
Know related work that belongs here? Suggest a link
Something wrong on this page? Tell the editor