Reproduce “Propagating the prior from shallow to deep with a pre-trained velocity-model Generative Transformer network” 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

Randy Harsuko, Shijun Cheng, Tariq Alkhalifah

VelocityGPT, a framework from KAUST, employs autoregressive generative Transformers for synthesizing seismic velocity models by propagating shallow subsurface information to deeper layers. The model effectively integrates diverse prior knowledge like geological classes, well data, and structural images, enabling the generation of geologically consistent and scalable velocity models.

Key insights

  • Autoregressive Transformer architectures can effectively capture and propagate the natural top-down information flow inherent in seismic data for velocity model generation.
  • Integrating diverse prior knowledge, such as geological classes, well-log velocities, and seismic structural information, significantly enhances the geological plausibility and constraint of generated models.
  • The autoregressive nature enables dynamic, spatially controlled generation, allowing for the injection of high-resolution zones or specific geological features at various depths.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.