The artificial intelligence (AI) method was used by SNAD researchers to find eleven previously undiscovered space anomalies.
In 2018, the team inspected the digital images of the Northern sky acquired with the help of a k-D tree to identify space anomalies via the “nearest neighbor” method. The research was then automated by using machine learning algorithms.
Surveying the sky with AI
There has been a vast increase in astronomical discoveries in recent years because of large-scale astronomical surveys. For instance, the Zwicky Transient Facility, surveys the Northern sky with a wide-field view camera, generating 1.4 TB of data per night of observation and a catalog containing billions of objects.
However, manually processing such massive amounts of data is prohibitively expensive and time-consuming. To address this, the SNAD team, which included researchers from Russia, France, and the United States, collaborated to develop an automated process.
When studying astronomical objects, scientists look at their light curves, which show how an object’s brightness changes over time. Scientists first detect a flash of light in the sky and then track its evolution to see if it grows brighter, weakens, or disappears.
The researchers examined a million real light curves from the ZTF’s 2018 catalog as well as seven simulated live curve models of the objects under consideration. They tracked 40 parameters, including an object’s brightness amplitude and timeframe.
We detailed our simulation properties with a set of characteristics anticipated to be monitored in real astronomical bodies, explained Konstantin Malanchev who co-authored the paper and is also a postdoc at Illinois University at Urbana-Champaign.
We were surveying for the following in a dataset of roughly a million objects. They include:
- Super-powerful supernovae
- Type Ia supernovae
- Type II supernovae, and
- Tidal disruption events
Such objects are referred to as space anomalies. They are either extremely rare, have little-known properties, or appear interesting enough to warrant further investigation.
Following that, the team used the k-D tree algorithm to compare light curve data from real objects to simulations. The k-D tree algorithm is a geometric data structure to divide space into lesser parts by cutting it with hyperplanes, planes, lines, or points. In the seven simulations, the algorithm was utilized to narrow the search range when surveying for real objects with similar properties to this one.
11 new space anomalies discovered
The researchers found 15 nearest neighbors (real objects from the ZTF database) for each simulation, for a total of 105 matches, which were then visually inspected for space anomalies. The manual verification process confirmed 11 space anomalies, seven of which were supernova candidates and four of which were active galactic nuclei candidates with the potential for tidal disruption events.
“This is a very good result,” explained Maria Pruzhinskaya, the paper’s co-author as well as a research fellow at the Sternberg Astronomical Institute. In addition to the previously discovered rare objects, we were able to detect several new ones that astronomers had previously missed. This means that existing search algorithms can be improved to prevent such objects from being missed.
The study shows that the method is highly efficient and simple to implement. Furthermore, the method is universal and can be used to find any astronomical object, not simply rare types of supernovae.
Matvey Kornilov, Associate Professor of Physics at HSE University, concluded, Astronomical and astrophysical phenomena that have yet to be discovered are, in fact, anomalies. Their observed manifestations are anticipated to differ from known object properties. In the future, we will try to discover new classes of objects using our method.
