Efficient physics-informed neural networks using hash encoding

Xinquan Huang, Tariq Alkhalifah

Abstract: Highlights•We propose an efficient physics-informed neural network by means of hash encoding.•We use FD method to obtain the first- and second-order derivatives and avoid the influence of discontinuous derivatives on AD.•We validate our method on the three PDE boundary value problems and achieve 10-fold acceleration in PINNs training.