Morephy-Net: An Evolutionary Multi-objective Optimization for Replica-Exchange-based Physics-informed Neural Operator Learning Networks
New neural operator model tackles noisy data and uncertainty quantification for complex physics simulations.
Deep Dive
Researchers Binghang Lu, Changhong Mou, and Guang Lin developed Morephy-Net, an evolutionary multi-objective optimization framework for physics-informed neural operator learning. It treats data and physics losses as separate objectives, uses replica-exchange sampling for stability, and provides Bayesian uncertainty quantification. The model shows improved accuracy and noise robustness over baselines like DeepONets when solving parametric partial differential equations (PDEs) for forward prediction and inverse problems.
Why It Matters
Enables more reliable AI simulation of complex physical systems with noisy or sparse real-world data.