The face of astronomy is changing. While narrow-field point-and-shoot astronomy remains important (JWST anyone?), large-scale wide-field surveys are expected to be the driving force behind discoveries in the coming decades, especially with the advent of machine learning. It has been.
A recently developed machine learning program called ASTRONOMALY scanned nearly 4 million galaxy images from the Dark Energy Camera Legacy Survey (DECaLS) and found 1635 images, including 18 unidentified sources with “highly unusual morphologies.” I found a number of abnormalities. This is a sign of things to come, where human-software partnerships can do better observational science than either could do alone.
Surveying telescopes have long been part of astronomers’ toolkits. What’s different in the 21st century is that humans can now generate incredibly large amounts of data, far more than we could mine and examine on our own. For example, the upcoming Vera Rubin Observatory will create 20 terabytes of data each night (60 petabytes over 10 years) and could eventually provide “32 trillion observations of 20 billion galaxies.” It is expected.
It would take humans decades to funnel all that data. AI can do it much faster.
Most previous anomaly detection programs were trained on a test dataset, teaching the algorithm to look for specific phenomena. A limitation of these programs is that they tend to detect many anomalies of the same type rather than entirely new anomalies.
ASTRONOMALY instead runs “unsupervised” and detects new kinds of outliers, the kind that excites astronomers, such as gravitational lensing, galaxy mergers, strange redshift patterns, and other oddities. can be found. However, ASTRONOMALY performs best when it employs a form of active learning that utilizes input from humans to correct mistakes. Incorporating this feedback into your search will give you better results.
The best part is that astronomers only need a few hours.
In a recent preprint paper, astronomers tested ASTRONOMALY on a larger dataset than ever before and demonstrated that it can operate at scale. After feeding the program a huge amount of her DECaLS data, I tested several different algorithms. The results showed that an unsupervised method powered by active learning input from humans provided the best output of inherent anomalies.
According to the researchers, the most interesting anomalies include “a ring galaxy exhibiting strange colors and morphology, a half-red and half-blue source, a potentially powerful lens system with a pair of sources acting as lenses, and several The known groups of interactions, and several sources of information, are interacting or coincidental.”
One of the mysterious objects emits radio radiation, which may be explained by the presence of quasars, but the galaxy also has a ring feature, which could be a rare red-ringed galaxy or gravitational lens. Another anomaly appears to be a ring-shaped starburst galaxy with either tidal tails or colliding companion galaxies.
All of these rare objects would have been missed without the active learning algorithm. The results promise exciting new discoveries in the very near future.
However, in this new era of vast datasets, there is still one challenge to overcome. It’s data transfer.
“One of the main challenges we experienced was transferring data from the host server to the local computer, which took several weeks,” the researchers said. What solutions did they propose? In the future, it makes more sense to bring computing power to the host observatory than to take data offsite.
learn more:
Verlon Etzebeth, Michel Rochner, Mike Walmsley, Margherita Grespin. “Large-scale astronomy: searching for anomalies among 4 million galaxies” ArXiv preprint.