Academic

Improving Full Waveform Inversion in Large Model Era

arXiv:2603.00377v1 Announce Type: new Abstract: Full Waveform Inversion (FWI) is a highly nonlinear and ill-posed problem that aims to recover subsurface velocity maps from surface-recorded seismic waveforms data. Existing data-driven FWI typically uses small models, as available datasets have limited volume, geological diversity, and spatial extent, leading to substantial concerns about overfitting. Although they perform well on synthetic datasets, current methods fail to generalize to more realistic geological structures. In this work, we show that a model trained entirely on simulated and relatively simple data can generalize remarkably well to challenging and unseen geological benchmarks. We provide a working recipe that tames a billion-parameter model for FWI through coordinated scaling across three axes: model capacity, data diversity, and training strategy. Our model achieves state-of-the-art performance on OpenFWI and significantly narrows the generalization gap in data-driven

Y
Yinan Feng, Peng Jin, Yuzhe Guo, Yinpeng Chen, Youzuo Lin · · 1 min read · 19 views
#cs.LG

arXiv:2603.00377v1 Announce Type: new Abstract: Full Waveform Inversion (FWI) is a highly nonlinear and ill-posed problem that aims to recover subsurface velocity maps from surface-recorded seismic waveforms data. Existing data-driven FWI typically uses small models, as available datasets have limited volume, geological diversity, and spatial extent, leading to substantial concerns about overfitting. Although they perform well on synthetic datasets, current methods fail to generalize to more realistic geological structures. In this work, we show that a model trained entirely on simulated and relatively simple data can generalize remarkably well to challenging and unseen geological benchmarks. We provide a working recipe that tames a billion-parameter model for FWI through coordinated scaling across three axes: model capacity, data diversity, and training strategy. Our model achieves state-of-the-art performance on OpenFWI and significantly narrows the generalization gap in data-driven FWI. Across six challenging geophysical benchmarks, including Marmousi, 2D SEG/EAGE Salt and Overthrust, 2004 BP, Sigsbee, and SEAM Phase I, it infers complex structures absent from the training set and delivers significant performance improvements (SSIM from 0.5844 to 0.7669). Overall, our results demonstrate that with an appropriate scaling strategy, large models trained on simple synthetic data can achieve substantial generalization to more complex and realistic geological structures.

Executive Summary

This study presents a novel approach to improving Full Waveform Inversion (FWI) in large model era by addressing the challenges of overfitting, limited dataset, and generalization to realistic geological structures. The authors propose a scaling strategy that balances model capacity, data diversity, and training strategy, allowing a billion-parameter model to generalize remarkably well to challenging geophysical benchmarks. The results demonstrate state-of-the-art performance on OpenFWI and significant improvements in SSIM across six benchmarks. This breakthrough has far-reaching implications for seismic imaging and subsurface mapping in the oil and gas industry.

Key Points

  • ▸ The authors propose a scaling strategy to balance model capacity, data diversity, and training strategy for FWI.
  • ▸ The approach enables a billion-parameter model to generalize well to challenging geophysical benchmarks.
  • ▸ The results demonstrate state-of-the-art performance on OpenFWI and significant improvements in SSIM across six benchmarks.

Merits

Generalization to Complex Geological Structures

The approach enables large models to generalize well to complex and realistic geological structures, addressing a significant limitation of current FWI methods.

Demerits

Dependence on Simulated Data

The approach relies heavily on simulated data, which may not accurately reflect real-world geological complexities and variability.

Expert Commentary

The study's innovative approach to scaling FWI models and its impressive results have significant implications for the oil and gas industry. However, the dependence on simulated data remains a concern, and further research is needed to validate the approach using real-world data. The study's findings also highlight the importance of balancing model capacity, data diversity, and training strategy in achieving generalization to complex geological structures. Overall, the study makes a valuable contribution to the field of seismic imaging and FWI.

Recommendations

  • ✓ Future research should focus on validating the approach using real-world data and exploring its application to other seismic imaging problems.
  • ✓ The oil and gas industry should consider adopting the scaling strategy and approach to generalization in their seismic imaging practices to improve efficiency and accuracy.

Sources

Related Articles