A Machine Learning Algorithm Finds its First Supernova

Plenty of recent mainstream news articles have touted AI’s ability to assist in the process of scientific discovery. But most of them predicted that it could take years or even decades to see the full effect. Astronomy seems ahead of the curve, though, with the announcement of a new AI system developed by researchers at Northwestern University and elsewhere that can now autonomously detect and classify supernovae.

The AI, or more accurately, the machine learning algorithm, which again in mainstream media is commonly confused for AI, is known as the Bright Transient Survey Bot (BTSbot). UT could not confirm whether the software was named after the ubiquitous Korean boy band. However, that seems a stretch, given the very different nature of their work.

BTSbot’s work is simple – crawling through massive databases, looking for a bright point of light that wasn’t there before. Those bright points represent one of the most fantastical of all astronomical phenomena – supernovae. And typically, it’s a herculean manual effort to find them.

https://www.youtube.com/watch?v=OEZu3pXAFWY&pp=ygUVZnJhc2VyIGNhaW4gc3VwZXJub3Zh Fraser explains why one of the brightest supernovas ever happened.

The developers of BTSbot, which include Adam Miller, the lead scientist and a professor of physics and astronomy at Northwestern’s Weinberg College of Arts and Sciences, and Christoffer Fremling, an astronomer at Caltech, estimate that humans have spent over 2,000 hours manually checking data collected by the Zwicky Transient Facility (ZTF) and other specialized supernovae hunters over the past six years.

That’s more than an entire year’s worth of working effort and could very clearly be put to more productive (not to mention more exciting) use, especially for the graduate students who are undoubtedly subjected to hundreds of hours scrolling through endless images of the night sky looking for a single bright point of light.

Luckily, BTSbot doesn’t get bored, so it is well-placed to do that trawling. But it’s also well placed to integrate into some automated infrastructure already in the supernovae hunting community. To understand how it’s probably easiest to talk through BTSbot’s first discovery of a supernova – though admittedly, the supernova it found, known as SN2023tyk, had already been found via the traditional manual search as well.

https://www.youtube.com/watch?v=RZKiWGT4Za4&pp=ygUVZnJhc2VyIGNhaW4gc3VwZXJub3Zh Don’t know what makes a supernova super? Fraser is here to explain.

ZTF first detected SN2023tyk on October 3rd, while BTSbot collected and analyzed its data in real-time. On October 5th, the algorithm announced it had found something, but what it did next shows how important integration is to modern scientific research.

Instead of immediately simply presenting its results to its parent operator, it sought to confirm the existence of the supernova and get some additional spectral data on it. To do this, it elicited the help of the SED Machine robotic telescope at the Palomar observatory. After collecting further data from the SED Machine, BTSbot passed that new data to another automated program – SNIascore, developed at Caltech by Dr. Fremling. 

SNIascore’s specialty is determining what kind of supernova an explosion was. It analyzes the spectra provided by both ZTF and Palomar and compares them to known values of other types of supernovae. It is not terribly tricky statistically, but it is still an impressive process when combined with the automated capture and analysis steps that came before it.

https://www.youtube.com/watch?v=L1otYvmqp6w&list=PLbJ42wpShvmkClWX_TrxFgNlP-p8-18cF&index=6&pp=iAQB It will be much harder for the universe to do anything without us looking with the advent of AI skywatchers.

The result of all this collaboration was a determination that SN2023tyk was a Type Ia supernova, which happens when a binary partner completely explodes. Dr. Miller and his colleagues announced the findings a few days later, on October 7th, and were quite pleased with the speed and accuracy with which BTSbot worked.

Admittedly, it is still early days for automated AI-assisted astronomy. But saving a full work year from aspiring students or fully-vested professors seems like a worthy goal. It’s only a matter of time before BTSbot becomes even more integrated into the astronomical community. And with that increased data input, who knows what it might find?

Learn More:Northwestern – First supernova detected, confirmed, classified and shared by AIUT – A new Kind of Supernova has Been DiscoveredUT – There’s a New Supernova in a Familiar Galaxy. You Can See it in a Small TelescopeUT – Supernovae Were Discovered in all These Galaxies

Lead Image:A Deep Field image of the galaxy containing the SN2023tyk supernova that BTSbot found.Credit – Legacy Surveys / D. Lang (Perimeter Institute) for Legacy Surveys layers and unWISE / NASA/JPL-Caltech / D. Lang (Perimeter Institute)

Source: https://www.universetoday.com/articles/a-machine-learning-algorithm-finds-its-first-supernova