Researchers are teaching computers to science more and more every year, it seems. Machine learning is the “in” field these days, and in new work coming from the Center for Astrophysics at Harvard, scientists have taught an AI to identify and classify supernovae.

Using a data set of 500 novae with spectra from the Pan-STARRS1 Medium Deep Sky Survey, computers can now be used to find supernovae in observations that don’t have spectra. And it turns out that there are a lot of observations without spectra. So by using this existing data set of actual observations and not just simulated data, the researchers created an AI that can identify supernovae with stunning accuracy.

The Pan-STARRS1 Medium Deep Sky Survey had 2,500 known supernovae in its catalog, and since only 500 of those had spectra, that data set could be used to train the AI to answer specific questions. Edo Berger, an astronomer at the CfA explained, “The machine learning looks for a correlation with the original 500 spectroscopic labels. We ask it to compare the supernovae in different categories: color, rate of evolution, or brightness. By feeding it real existing knowledge, it leads to the highest accuracy, between 80- and 90-percent.”

This technique is going to prove even more useful once the Vera Rubin Observatory is online because while the Large Synoptic Survey Telescope [Ed. note: This name is defunct. LSST is the Legacy Survey of Space and Time] will surely discover a ton of new supernovae, the percentage we’ll have spectra for will not increase. Now scientists will be able to use all the other components of the observations to classify all those new novae we find. The corresponding papers for this work are published in The Astrophysical Journal.

More Information

Center for Astrophysics press release 

“SuperRAENN: A semi-supervised supernova photometric classification pipeline trained on Pan-STARRS1 Medium Deep Survey supernovae,” A. Villar et al, 2020 December 17, The Astrophysical Journal (preprint on arxiv.org)

“Photometric Classification of 2315 Pan-STARRS1 Supernovae with Superphot“, G. Hosseinzadeh et al, 2020 December 17, The Astrophysical Journal (preprint on arxiv.org)