Machine Learning Validates Unrecognized Transient Astronomical Phenomena in Historical Observatory Images
By
solarist
Right out the toaster. Reliable, with some real depth.
Summary
This research paper uses machine learning to validate the existence of previously unrecognized transient astronomical phenomena in historical observatory images. The study addresses debates that these transient point sources (appearing and vanishing over short timescales) are simply plate defects. An ML model trained on 250 expert-classified image pairs achieved good discrimination (AUC=0.81). Applied to 107,875 previously-identified transients, the model found that transient counts were significantly elevated within one day of nuclear testing (p=.024) and that the highest-probability transients were even more strongly associated with nuclear testing windows (p<.0001). A significant shadow deficit (fewer transients in Earth's shadow) was also confirmed (p<.0001), with the effect strongest in the highest-probability transients. The results strongly support the existence of a real, unrecognized population of transient objects in historical astronomical plates.
Key quotes
· 5 pulledResults strongly support existence of an unrecognized population of transient objects in historical astronomical plates warranting further study.
After controlling for ML-identified artifacts, transient counts were significantly elevated for dates within a nuclear window (p=.024)
The shadow deficit was significant (p<.0001) and largest in the highest probability transients relative to lower probability transients (p=.003)
The model demonstrated good discrimination (out-of-fold AUC=0.81; sensitivity=0.71, specificity=0.71)
These findings remain debated with some arguing that transients identified via existing automated pipelines are simply plate defects.
