Four examples of simulated FRBs injected into real data from the CHIME Pathfinder survey. By combining thousands of these true positives with known false positives, we have built a large labeled training set. The three panels in each subfigure are the frequency–time intensity array of the dedispersed pulse, the frequency-averaged pulse profile, and the DM–time intensity array (top to bottom). Combinations of these data products can be used as inputs to the multi-input neural net described in Section 3.