Paper submitted to the 50th IPA Convention & Exhibition, by Jyoti Kumar, Øystein Korsmo and Nizar Chemingui (TGS)
Summary
This paper looks at how deep learning can improve the initial Full Waveform Inversion (FWI) velocity model and reduce the need for traditional tomography in velocity model building (VMB), which can be time-consuming. The main goal is to make the VMB process faster and more efficient, but deep learning also opens new possibilities — allowing better data handling, quicker convergence, and improved performance in areas where ray-based methods struggle, especially in complex, high-contrast environments.