This approach could save astronomers a lot of time searching for life in the universe.
Introduction: The University of Warwick recently used an AI algorithm that enables AI to instantly identify planetary transit signals during observation through big data learning, and automatically identify exoplanets, and 50 new exoplanets have been discovered using this algorithm.
In 1995, astronomers Aleksander Wolszczan and Dale Frail were observing a distant pulsar when they noticed that the signal frequency of its pulsed light appeared to skip a beat.
This out-of-rhythm pulse of light led to the discovery of the first exoplanet, called 51 Pegasus. Since then, the industry of finding unusual and distant planets (like Earth's) orbiting stars other than the Sun has boomed, and to date, more than 4,000 exoplanets have been confirmed.
With space missions that are being searched in the universe, the research continues. To speed up the process, a team of astronomers came up with a new machine learning algorithm. This algorithm can rule out false positives and confirm the true exoplanets.
With the help of an artificial intelligence system, a team of astronomers from the University of Warwick confirmed the existence of 50 new planets.
The new technique was detailed in a study published this week in the Monthly News of the Royal Astronomical Society.
Some examples of exoplanets recently discovered by NASA's Kepler mission. NASA/Ames/JPL-Caltech
The two main missions behind the ongoing exoplanet excavation mission are NASA's Kepler mission and the Transiting Exoplanet Survey Satellite (TESS) mission. The Kepler telescope takes close-up shots of stars to see if there are planets orbiting in habitable zones, while TESS investigates the 200,000 brightest stars in the sky.
When a planet passes in front of its host star, it causes the star's light to drop slightly. Exoplanet hunters look for this descent as a clue that there may be a planet orbiting the star on the surface. But after the initial probe, astronomers need to confirm that it was actually a planet that caused the change in light.
There are more than 3,000 such signals waiting to be identified as planets, rather than misexamined.
David Armstrong, a professor in the Department of Physics at the University of Warwick and the main initiator behind the new study, wanted to compile an algorithm that would help astronomers sift through the data more quickly.
"We wanted to develop this algorithm to handle the task in progress." Armstrong told Astronomy Online, "It's much faster than before, and we can apply him to more candidate planets."
The researchers trained the algorithm to distinguish between real and false positives by entering a large number of confirmed planets and false positives. By comparing it with the data in the database, the machine learning system learned how to distinguish between the two categories.
The researchers then used the algorithm in a dataset of thousands of unconfirmed planets, including those found in NASA's Kepler program.
The algorithm is capable of identifying 50 new planets from the dataset. These planets are giants the size of Neptune to earth-like giants, and their orbits range from one day to 200 days.
Armstrong said: "What we do differs from other machine learning techniques is that we try to identify which are new planets that are statistically likely to be planets based on probabilities, rather than just sorting them. ”
Previous machine learning algorithms would have sorted exoplanets based on how likely they are to become planets, but this new algorithmic system determines the probability that each new planet will become a planet in itself.
"The 50 new planets exported exceed the threshold of a 99 percent probability of being judged to be planets." Armstrong said, "These are planets that cross this line. ”
As new discoveries of exoplanets continue, researchers will continue to provide the algorithmic system with more candidate planets, and it will gradually improve its skills. The faster we can identify real planets from false ones, the faster we can track those planets in the universe and explore habitability.
To date, more than 30% of the 4,000 known exoplanets have been "verified" to be discovered, calculating the statistical likelihood of non-planetary scenarios caused by misexamination (FP: Treating Non-Planets as Planetary Conditions). For the vast majority of validated planets, the calculations are performed using vespa (open source big data services engine) computation programs (Morton codes or other algorithms. 2016).
Regardless of the strengths and weaknesses of vespa, we very much want the catalog of known planets not to rely on a single approach. We demonstrated how to use machine learning algorithms, specifically gaussian process classifiers (GPCs) enhanced by other models, to perform probabilistic planetary validation that includes a priori probabilities for possible FP scenarios.
When the confirmed planets are separated from misinspective conditions (FPs) in the Kepler Threshold Crossing Events (TCE) catalog, GPC can obtain an average logarithmic loss of 0.54 per sample. Once the applicable review metrics have been calculated, our model can validate thousands of invisible candidates in seconds. And it can be adapted to collaborate with active transiting exoplanet survey satellite (TESS) missions, where a large number of observed targets require the use of automated algorithms.
We discussed the limitations and considerations of this approach, and after considering possible failure modes, revalidated the 50 Kepler candidates about the planet, validating them for plausibility testing by using vespa's latest stellar information confirmation. Regarding the difference with vespa leading to the emergence of many other candidate planets, this is often beneficial to solve our model. With these issues in mind, we caution against using a single method of planetary verification using the two methods until the differences are fully understood.
BY: inverse
If there is any infringement of the relevant content, please contact the author to delete it after the work is published
Please also obtain authorization to reprint, and pay attention to maintaining completeness and indicating source