A modified physics-informed neural network with positional encoding (Huang, X., et al, 2021)

To address common issues of physics-informed neural network (PINN) that the predicted solutions were somewhat smooth and the convergence of the training was slow, we propose a modified PINN using sinusoidal activation functions and positional encoding, aiming to accelerate the convergence and fit better. We transform the scalar input coordinate parameters using positional encoding into high-dimensional embedded vectors and train a fully-connected neural network to predict the real and imaginary parts of the scattered wavefield. Compared to the commonly used PINN, the proposed modified PINN using positional encoding exhibits notable superiority in terms of convergence and accuracy.

Figures_for_Research9_Xinquan

References

 

Huang X., Alkhalifah T., and Song C., 2021, "A modified physics-informed neural network with positional encoding", SEG Technical Program Expanded Abstracts: 2480-2484. https://doi.org/10.1190/segam2021-3584127.1

 

Huang X., Alkhalifah T., and Song C., 2021, "A modified physics-informed neural network with positional encoding", SEG Technical Program Expanded Abstracts: 2480-2484. https://doi.org/10.1190/segam2021-3584127.1