We propose a meta-learning technique for offline discovery of physics-informed neural network (PINN) loss functions. We extend earlier works on meta-learning, and develop a gradient-based meta-learning algorithm for addressing diverse task distributions based on parametrized partial differential equations...
and stitching the chunks together to approximate the full field response. We emphasize that FDFD simulations were never carried out to train the PINN, and we only minimized the residual of the PDE itself to train the network. The PINN training took approximately 2 hours on our machine. In our...