Bob McDonald's blog: Machine learning can analyze large datasets and pick out details missed by humans
Astronomers are turning to artificial intelligence and machine learning to help find new planets forming around other stars and possible signals from intelligent civilizations elsewhere in the galaxy.
The flood of new data streaming down from the James Webb Space Telescope and other large instruments on the ground is overwhelming the humans who go through the laborious and time-consuming process of looking for interesting patterns and phenomena.
Intelligent machines, on the other hand, can be very good at recognizing patterns among large amounts of data and may be able to pick out subtle signals that humans might miss.
Researchers at the University of Georgia have experimented with a machine learning system to teach the algorithmto spot newly forming planets among dust rings surrounding other stars.
Solar systems, including our own, are born out of large clouds of gas and dust that are pulled together by gravity. Over time, most of the material falls to the centre where it becomes a star, while planets form within a leftover disc of dust that surrounds the star.
Many of these discs have been spotted but finding newly forming planets hiding among the dust is a challenge. The dust tends to be thick so young planets can only be detected by the disturbances they make as they pass through the dust, similar to the wake left behind a boat moving through water.
In a proof-of-concept study, the scientists used computer simulations of young solar systems with new planets forming in them to teach the machines how to recognize the patterns of disturbance.
They then tested the algorithm out on real astronomical data gathered by the Atacama Large Millimeter Array of radio telescopes in Chile, which has been doing surveys of stellar disks. In particular, they let the algorithm look at data where manual examination of the data had spotted signs of a planet, and the AI identified the same systems with better than 90 per cent accuracy.
These programs are much faster than manual methods and could lead to the discovery of many new worlds. The more of these so-called proto-planets that are found, the more details about planetary formation are revealed.
Tuning into alien transmissions
Another group at the Search For Extraterrestrial Intelligence (SETI) Institute is also using artificial intelligence to look at data from radio telescopes that have been scanning the skies looking for signals that could be coming from other civilizations in our Milky Way.
Space is a noisy place, with radio emissions coming from a wide variety of objects, such as exploding stars, black holes, pulsars, radio galaxies as well as the cacophony of radio noise generated by our technology here on Earth. Picking out a signal that could be from an alien civilization among all that noise is a needle in a haystack problem that has so far yielded no results, despite decades of searching.
The latest study looked at data from 820 objects in space representing 480 hours of searching as part of the Breakthrough Listen initiative. Eight "signals of interest" were found that had not been detected in the data by other methods. Unfortunately when the source of those radio signals was re-observed, the signals were gone, so no alien source could be confirmed.
So the search for other intelligent civilizations in space continues.
It will soon be up to intelligent machines here on Earth to search through that data to find other new worlds and possibly alien intelligences calling out across the cosmos.
Credit belongs to : www.cbc.ca