Large AI model trained on synthetic data achieves SOTA in seismic inversion

Researchers have developed a novel approach to Full Waveform Inversion (FWI) by leveraging a large, billion-parameter model trained on simulated data. This method addresses the common issue of overfitting in data-driven FWI, where limited datasets typically hinder generalization to real-world geological structures. The proposed strategy involves coordinated scaling across model capacity, data diversity, and training techniques, enabling the model to achieve state-of-the-art performance on benchmarks like OpenFWI and significantly improve generalization capabilities. AI

IMPACT Demonstrates that large models trained on synthetic data can generalize to complex real-world problems, potentially advancing AI applications in geophysics.

RANK_REASON Academic paper detailing a new methodology for seismic inversion using large AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• 2D SEG/EAGE Salt and Overthrust
• OpenFWI
• SEAM Phase I
• Sigsbee
• Yinan Feng

• paper
• model release

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