NSGA-PINN: A Multi-Objective Optimization Method for Physics-Informed Neural Network Training
Published in Algorithms, 2023
NSGA-PINN recasts PINN training as multi-objective optimization, applying an NSGA-II evolutionary scheme to jointly minimize the data-fitting loss and the physics-residual loss. The Pareto-based approach mitigates the loss-balancing pathologies that destabilize gradient-based PINN training and improves accuracy on both forward and inverse PDE problems.
Recommended citation: Lu, B., Moya, C., & Lin, G. (2023). "NSGA-PINN: A Multi-Objective Optimization Method for Physics-Informed Neural Network Training." Algorithms, 16(4), 194.
Download Paper