Poseidon
Efficient Foundation Models for PDEs
Paper title Poseidon: Efficient Foundation Models for PDEs
Paper arXiv 2405.19101PDF
A multiscale vision-transformer operator with time-conditioned layer norms, so the solution can be asked for at any time. Pretraining uses every pair of snapshots in a trajectory, which multiplies the training examples, on compressible Euler and incompressible Navier–Stokes data, before fine-tuning on a new task.
- Published
- NeurIPS
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
More papers
All papers- PretrainingUniversal latent spaces — Self-Supervised Learning of Universal Latent Spaces for Operator Learning
- PretrainingCross-domain pretraining — Cross-domain pretraining for steady neural CFD surrogates
- PretrainingGeometry meets physics — Data-efficient pre-training for unstructured neural PDE solvers