Neural Solver

Method Stack

• PINN for PDE-constrained learning
• Neural Operators for cross-resolution generalization
• Hybrid solvers combining numerical methods and neural models

Engineering Focus

• Training stability (loss balancing, sampling)
• Parallel efficiency (data/model parallel)
• Validation against classical numerical solvers