Morphy-Net: Evolutionary Multi-Objective Optimization for Replica-Exchange-Based Physics-Informed Neural Operator Learning Networks
Published in Computer Methods in Applied Mechanics and Engineering (CMAME), 2026
Morphy-Net treats neural-operator training as a multi-objective, multi-population search: an evolutionary outer loop balances data, residual, and regularization objectives, while a replica-exchange inner loop swaps parameters across temperatures to escape the sharp local minima that plague PINN/DeepONet training. The combination yields Pareto fronts of operator surrogates with markedly improved generalization across PDE families.
Accepted by Computer Methods in Applied Mechanics and Engineering (CMAME), 2026.
Recommended citation: Lu, B., Mou, C., & Lin, G. (2026). "Morphy-Net: Evolutionary Multi-Objective Optimization for Replica-Exchange-Based Physics-Informed Neural Operator Learning Networks." Computer Methods in Applied Mechanics and Engineering (accepted).
Download Paper