New PSG method enhances seismic inversion with diffusion models

Researchers have developed a new method called Physical-State-Guided Diffusion Sampling (PSG) to improve full-waveform inversion (FWI) for subsurface velocity estimation. PSG couples a persistent physical velocity with a diffusion prior through a Gaussian bridge, refining the physical state via waveform fitting. This approach separates wave-equation and denoiser gradients, allowing for better initialization and optimization history. PSG has demonstrated superior performance over existing methods on various OpenFWI families, even with noisy or incomplete seismic data, and successfully recovered complex geological structures in larger models like Marmousi, Overthrust, and BP2004 Salt without retraining. AI

IMPACT This method could improve the accuracy and efficiency of subsurface imaging for resource exploration and geological studies.

RANK_REASON The cluster contains a research paper detailing a new method for seismic data analysis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• BP2004 Salt
• Gaussian bridge
• Hugging Face
• Marmousi model
• OpenFWI
• Overthrust
• Physical-State-Guided Diffusion Sampling

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →