Search papers, labs, and topics across Lattice.
This paper introduces a physics-guided diffusion framework for full-waveform inversion (FWI) that leverages a score-based generative prior and likelihood guidance from wave-equation simulations. The method uses a transport-based data-consistency potential with wavefield enhancement to improve robustness against amplitude and phase errors. A preconditioned guided reverse-diffusion scheme is proposed to adapt guidance strength during inference, leading to more stable and effective data consistency.
Diffusion models can now solve full-waveform inversion problems more robustly than traditional optimization or standard diffusion posterior sampling, even with noisy data.
We develop a robust physics-guided diffusion framework for full-waveform inversion that combines a score-based generative prior with likelihood guidance computed through wave-equation simulations. We adopt a transport-based data-consistency potential (Wasserstein-2), incorporating wavefield enhancement via bounded weighting and observation-dependent normalization, thereby improving robustness to amplitude imbalance and time/phase misalignment. On the inference side, we introduce a preconditioned guided reverse-diffusion scheme that adapts the guidance strength and spatial scaling throughout the reverse-time dynamics, yielding a more stable and effective data-consistency guidance step than standard diffusion posterior sampling (DPS). Numerical experiments on OpenFWI datasets demonstrate improved reconstruction quality over deterministic optimization baselines and standard DPS under comparable computational budgets.
