Code
SciMBA
Scientific Machine Learning library for solving parametric PDEs
SciMBA is an open-source Python library implementing Scientific Machine Learning (SciML) methods for PDE problems. It provides tools for hybrid numerical methods and supports a range of nonlinear approximation spaces including neural networks, low-rank approximations, and kernel methods.
Key features:
- Physics-Informed Neural Networks (PINNs) and Deep Ritz methods
- Neural Galerkin schemes
- Natural gradient optimization
- Support for complex geometries (level-set techniques), elliptic, time-dependent, and kinetic PDEs
- Two backends:
scimba_torch(PyTorch) andscimba_jax(JAX)
I contribute to the development of SciMBA, with a focus on natural gradient methods and kernel-based techniques.
More projects coming soon.