New self-attention network enhances seismic velocity model building

Researchers have developed a novel multiscale self-attention summary network designed to improve subsurface velocity-model building for seismic imaging. This network maps high-dimensional common-image gather (CIG) volumes into compact conditioning embeddings, which are then used by a flow-matching model to infer plausible velocity fields. The approach demonstrates enhanced accuracy and reduced uncertainty in posterior velocity reconstructions compared to methods that directly use raw CIG data, with the multiscale attention design offering greater robustness to background-model mismatches. AI

IMPACT Introduces a novel deep learning architecture for geophysical data analysis, potentially improving accuracy and efficiency in subsurface imaging.

RANK_REASON The cluster contains a research paper detailing a new machine learning model for a specific scientific application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Gaussian source distribution
• Hugging Face
• Posterior Distribution
• Self-attention summary networks

• paper
• infra

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