Shijun Cheng, Mohammad H. Taufik, and Tariq Alkhalifah

 GitHub: https://github.com/DeepWave-KAUST/GNO-pub

This work proposes a Generative Neural Operator (GNO) for seismic scattered wavefield modeling, integrating a diffusion-based generative framework with physics-guided sampling. By training on diverse velocity models, frequencies, and source positions, the method learns the statistical distribution of scattered wavefields and quickly synthesizes accurate solutions at inference time. Incorporating wave-equation constraints at each sampling step enhances both convergence speed and physical fidelity. Numerical experiments show that the GNO outperforms a baseline Fourier Neural Operator, especially for out-of-distribution velocity models and frequencies.