A weak-form-centric differentiable finite element framework in JAX

Weak-form physics-informed learning for noisy scientific image dynamics. Two domain validations: Flare-PINN (solar flare forecasting) and FRAP (diffusion-coefficient recovery).

A hybrid deep learning framework combining physics-informed neural networks (PINNs) with magnetohydrodynamic (MHD) constraints for operational solar flare forecasting.