Human interpretation, machine learning and point processes for the identification of faults on seismic sections

@inproceedings{tatymoukati:hal-05157114,
 abstract = {This work presents a new way to integrate human interpretation to automatically downscale a machine learning-based fault likelihood image into a set of potential fault networks. The downscaling involves a statistical marked point process with interactions to simulate the fault network consistently with the likelihood image. The model parameters are inferred from reference fault networks as produced by experts, using an effective Bayesian inference method (the ABC Shadow algorithm). This opens interesting perspectives to reduce the gap between computational structural uncertainty quantification and human-based deterministic approaches, and to rigorously assess its impact on model-based subsurface forecasts.},
 address = {Toulouse, France},
 author = {Taty Moukati, Fabrice and Caumon, Guillaume and Stoica, Radu S.},
 booktitle = {{86th EAGE Annual Conference \& Exhibition}},
 hal_id = {hal-05157114},
 hal_version = {v1},
 month = {June},
 pages = {1-5},
 publisher = {{European Association of Geoscientists \& Engineers}},
 title = {{Human interpretation, machine learning and point processes for the identification of faults on seismic sections}},
 url = {https://hal.univ-lorraine.fr/hal-05157114},
 volume = {2025},
 year = {2025}
}