Geometry Meets Physics: Data-Efficient Pre-Training for Unstructured Neural PDE Solvers
What it is
This paper introduces a disk-data-free pre-training framework for neural surrogate models of Partial Differential Equations (PDEs) on unstructured 3D geometries. It uses geometry-driven shape descriptors for steady-state problems and online synthetic data generation for transient problems to avoid the high cost of large pre-computed datasets.
The problem it solves
Existing pre-training strategies for neural PDE solvers are neither compute- nor data-efficient, relying on massive, expensive-to-generate pre-computed datasets that limit scalability and generalization.
What you could use it for
- Pre-training neural emulators for complex 3D fluid dynamics
- Developing data-efficient surrogates for transient heat transfer simulations
- Improving generalization of PDE solvers in low-data regimes
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.