Physic Informed Neural Network
Physics-informed neural networks (PINNs) integrate physical laws, typically expressed as differential equations, into neural network training to solve complex scientific problems. Current research focuses on improving PINN accuracy and efficiency through architectural innovations like Fourier-based networks, Kolmogorov-Arnold networks, and wavelet-based approaches, as well as advanced optimization strategies such as dual cone gradient descent and DiffGrad. These advancements aim to overcome limitations in handling high-frequency solutions, complex geometries, and stiff equations, ultimately enhancing the applicability of PINNs across diverse scientific and engineering domains, including fluid dynamics, seismology, and materials science.
Papers
DEQGAN: Learning the Loss Function for PINNs with Generative Adversarial Networks
Blake Bullwinkel, Dylan Randle, Pavlos Protopapas, David Sondak
Bi-level Physics-Informed Neural Networks for PDE Constrained Optimization using Broyden's Hypergradients
Zhongkai Hao, Chengyang Ying, Hang Su, Jun Zhu, Jian Song, Ze Cheng