Original title: SNAD: enabling discovery in the era of big data
Maria Pruzhinskaya, Emille E. O. Ishida, Konstantin Malanchev, Anastasia Lavrukhina +7 · August 2026
Telescopes collect huge piles of sky survey data, making it hard to spot rare astronomical objects. The SNAD team built tools to help find these unusual objects using machine learning, which means computer programs that learn from data. They believe discovery works best when algorithms help human domain experts rather than working completely alone. To do this, they created the Coniferest Python library with active learningA machine learning method where the algorithm chooses which data points to ask a human to label, so it learns faster with fewer examples., a method where an algorithm asks a human expert for guidance. They also built the SNAD Transient Miner to locate specific changing objects and the SNAD Viewer web portal to inspect data on individual objects. Looking at Zwicky Transient FacilityA large astronomical survey that scans the sky for transient events using a wide-field camera. data, they found more than a hundred new supernova candidatesObjects that look like they might be supernovas but need more observation to confirm. and uncatalogued objects like red dwarf flares and young stellar objects.
Imagine you are a detective looking through thousands of security camera photos to find a rare clue. The computer flags suspicious ones, but it often misses subtle clues that only an experienced eye can spot. So you teach the computer by pointing out the ones you find, and it gets better at finding similar ones. In this paper, the astronomers act as the detective, and the machine learning system learns from their expertise to find rare astronomical objects in vast sky surveys.
Astronomical surveys produce so much data that automated systems can miss rare or unexpected objects. By keeping a human expert in the loop, this approach catches more of those discoveries—like new supernovas and stellar flares—that would otherwise stay hidden. It means we can find the unusual faster and more reliably.
The abstract does not specify limitations, but the approach depends on expert availability and may not scale to all surveys. The reported candidates require spectroscopic confirmation to be validated as genuine discoveries.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
SNAD: enabling discovery in the era of big data — Maria Pruzhinskaya, Emille E. O. Ishida, Konstantin Malanchev et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.