Mark Megahan
Ever since the first vague shadow of an extra-solar planetary object was observed around Beta Pictoris in 1984, astronomers have cataloged 4,569 exoplanets in total. Thanks to an artificial intelligence algorithm called ExoMiner, 301 brand new entries were added in one whack.
Exoplanets popping up everywhere
The human race has eternally longed to walk on the surface of a distant planet, orbiting some strange point of light. The law of averages alone suggests that there should be plenty of them to choose from.
Ever increasing technology has led to the discovery of exoplanets by the thousands but none, so far, have been discovered in the narrow “sweet spot” of perfect distance from a suitable star to support life as we know it.
That may change in the near future. Deep learning concepts have been applied to a new neural network called ExoMiner, “which has been running on the Pleiades, NASA’s supercomputer.”
It’s looking for habitable exoplanets and it’s better than humans at finding them. Much better. They haven’t finished adjusting the program yet and it already pegged 301 verified planets outside our own solar system.
After they fed in one dataset at a time, “ExoMiner has developed the capacity to distinguish between genuine exoplanets and false positives without the need for human scientists to sift through the datasets themselves.”
The program is teaching the astronomers a few things too. By looking at the data through the program’s eyes in debugging mode, scientists are better able to “determine the criteria for and properties of” the elusive targets.
Train the network
That helps them “train the neural network to identify these characteristics and decide what data fits them.”
First the program gives it a go, scientists check the work, and the “network learns from previously analyzed data where scientists have validated exoplanets and false positives.”
The team started with the data from NASA’s Kepler spacecraft. Next, they layered in the K2 follow-on mission. “With thousands of stars in the field of view of the Kepler space telescope and similar missions, all with the possibility of hosting exoplanets, it would take significantly longer for scientists to determine which of them do, compared to a neural network.” The programmers are proud of their new friend.
“Unlike other exoplanet-detecting machine learning programs, ExoMiner isn’t a black box – there is no mystery as to why it decides something is a planet or not.” According to Jon Jenkins with Ames Research Center, “We can easily explain which features in the data lead ExoMiner to reject or confirm a planet.”
Hamed Valizadegan, ExoMiner project lead, is an even bigger fan. “When ExoMiner says something is a planet, you can be sure it’s a planet. ExoMiner is highly accurate and in some ways more reliable than both existing machine classifiers and the human experts it’s meant to emulate because of the biases that come with human labeling.”
The search for exoplanets should soon pay off with one that we can live on. “Now that we’ve trained ExoMiner using Kepler data, with a little fine-tuning, we can transfer that learning to other missions, including TESS, which we’re currently working on,” Valizadegan notes. “There’s room to grow.”
2 Comments