Structure-Preserving Learning Improves Geometry Generalization in Neural PDEs
Learn how to improve geometry generalization in neural PDEs using structure-preserving learning, which provides real-time solutions to Partial Differential Equations while preserving physical conservation laws
- Implement General-Geometry Neural Whitney Forms (Geo-NeW) to jointly learn a differential operator and compatible reduced finite element spaces
- Use Finite Element Exterior Calculus to preserve physical conservation laws
- Apply transformer-based encoding to connect the underlying geometry and imposed boundary conditions to the solution
- Develop a novel parameterization of the constitutive model to ensure the existence and uniqueness of the solution
- Evaluate the performance of the model on out-of-distribution geometries and compare with conventional baselines
Researchers and engineers working on physics foundation models for science and engineering can benefit from this approach, as it improves generalization to unseen domains and provides state-of-the-art performance on several steady-state PDE benchmarks
💡 Structure-preserving learning can improve generalization to unseen domains by explicitly connecting the underlying geometry and imposed boundary conditions to the solution
🚀 Improve geometry generalization in neural PDEs with structure-preserving learning! 📈
Key Takeaways
Learn how to improve geometry generalization in neural PDEs using structure-preserving learning, which provides real-time solutions to Partial Differential Equations while preserving physical conservation laws
DeepCamp AI