LNN-PINN: Physics-Informed Neural Network with Liquid Residual Blocks

A new framework called LNN-PINN has been introduced by researchers, which utilizes a physics-informed neural network design that incorporates a liquid residual gating structure to improve prediction precision. This approach features a streamlined gating system in the hidden-layer mappings, while maintaining the same sampling, loss composition, and hyperparameters, guaranteeing that enhancements arise exclusively from architectural adjustments. In testing on four benchmark problems, LNN-PINN consistently demonstrated reductions in RMSE and MAE under the same training conditions, with absolute error plots validating the improvements in accuracy.