Astronomaly: Machine Learning Only Finds Rare Objects with an Astronomer in the Loop

Michelle Lochner defends partial automation in the face of the volumes of the SKA and the Vera C. Rubin Observatory: it is the interface between algorithm and expert that produces the discoveries.

Instruments that enter into service produce quantities of data beyond the scope of traditional analytical methods. The Square Kilometre Array and the Vera C. Rubin Observatory are at the top of this category. In a text from a conference invited to the IAU General Assembly, Michelle Lochner examines what machine learning really makes it possible to gain.

His observation nuances the idea of a fully automated scientific discovery. Excellent algorithms to process volume and handle previously manual tasks, but significant findings occur when active learning combines the machine with human expertise rather than without it.

Astronomaly, the public setting she presents, illustrates the approach: the system detects anomalies and recommends to the astronomer sources that deserve her attention, learning from her feedback to refine her suggestions. It is this cyber-human interface, more than computing power alone, that brings out classes of rare or unpublished objects in large surveys.