Reproduce “Fast ground penetrating radar dual-parameter full waveform inversion method accelerated by hybrid compilation of CUDA kernel function and PyTorch” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lei Liu, Chao Song, Liangsheng He, Silin Wang, Xuan Feng, Cai Liu
Researchers at Jilin University developed a Ground Penetrating Radar (GPR) dual-parameter full waveform inversion (FWI) method, achieving significant computational acceleration through a hybrid compilation of CUDA kernel functions and PyTorch. This framework accurately reconstructs subsurface dielectric permittivity and electrical conductivity, completing inversion tasks in minutes on synthetic and real GPR data.
Key insights
- A hybrid approach effectively combines the computational efficiency of custom CUDA kernels for numerical simulations with the flexible automatic differentiation capabilities and ecosystem of PyTorch for gradient-based optimization.
- Dual-parameter GPR FWI for permittivity and conductivity can be robustly performed by a two-stage inversion strategy, first focusing on permittivity and then simultaneously optimizing both parameters with careful learning rate scheduling.
- Keeping all computational data in GPU memory and minimizing data transfer overhead is crucial for achieving orders-of-magnitude speedups in complex geophysical inversion problems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.