Full-waveform inversion (FWI) is a widely used technique for reconstructing subsurface velocity models from seismic data. However, its practical performance is hampered by several known issues: the problem is often ill-conditioned, the solution is sensitive to the choice of update direction, and the computational cost can be prohibitive.

The new arXiv paper, titled "Full Waveform Inversion with a Learned Preconditioner," proposes to address these limitations by incorporating a learned preconditioner. The abstract indicates that the preconditioner is intended to improve the conditioning and guide the inversion process, though the specific architecture or training approach is not described in the available text.

It is worth noting that the abstract is cut off mid-sentence, so the full scope of the contribution—including any experimental results or comparisons—is not yet accessible from this source. Readers interested in the details will need to consult the full paper on arXiv.