We need to understand where faults are to do the following: (1) assess their seismic hazard, (2) explore for natural resources and (3) store CO₂ safely in the subsurface. Currently, we still map subsurface faults primarily by hand using seismic reflection data, i.e. acoustic images of the Earth. Mapping faults this way is difficult and time-consuming. Here, we show how to use deep learning to accelerate fault mapping and how to use networks or graphs to simplify fault analyses.