Demonstration of the interactive training procedure. In a “query,” a set of objects is shown to the “oracle” (the user), who provides an interest level score for each object (“1” for compact sources or pure noise, to “5” for highly unusual morphology). Images are normalized using asinh scaling with the brightness truncated to 90% of maximum. The active learning algorithm described in Section 6 is then applied, the objects are sorted by acquisition score (which prioritizes which objects will provide the algorithm with the most useful information), and the next set of objects are queried. Within just a few iterations, the algorithm hones in on high human-scoring sources.