- New algorithm for machine learning designed by astronomers and computer scientists University of Warwick confirms new exoplanets in telescope data
- Sky research finds thousands of planetary candidates, and astronomers must separate the true planets from false ones
- Algorithm was trained to distinguish between signs of real planets and false positives
- New technique is faster than previous techniques, can be automated, and improved with further training
Fifty potential planets have confirmed their existence through a new machine-learning algorithm developed by scientists at the University of Warwick.
For the first time, astronomers have used a process based on machine learning, a form of artificial intelligence, to analyze a sample of potential planets and determine which ones are real and which ‘false’, or false positives, calculate the probability of each. candidate to be a true planet.
Their results are reported in a new study published in the Monthly announcements from the Royal Astronomical Society, where they also perform the first large-scale comparison of such planetary validation techniques. Their conclusions make the case for using multiple validation techniques, including their machine-learning algorithm, as statistical future confirms exoplanet discoveries.
Many exoplanet studies are searching through vast amounts of telescope data for the signs of planets passing between the telescope and its star, known as transit. This results in a telltale dip in light from the star that detects the telescope, but it can also be caused by a binary star system, interference from an object in the background, or even slight errors in the camera. These false positives can be sifted through a planetary validation process.
Researchers from Warwick’s departments of physics and computer science, such as the Alan Turing Institute, built an algorithm based on machine learning that can separate real planets from false ones in the grand examples of thousands of candidates found by telescopic missions such as NASA‘s Kepler and TESS.
It was trained to recognize real planets with two great examples of confirmed planets and false positives of the now retired Kepler mission. The researchers then used the algorithm on a dataset of Kepler’s as yet unconfirmed planetary candidates, resulting in fifty new confirmed planets and the first to be validated by machine learning. Earlier machine learning techniques ranked candidates, but never determined the chance that a candidate was even a true planet, a mandatory step for planet validation.
Those fifty planets vary from worlds as large as Neptune to smaller than Earth, with jobs as long as 200 days to as little as one day. By confirming that these fifty planets are real, astronomers can now give priority to further observations with dedicated telescopes.
Dr. David Armstrong, of the Department of Physics at the University of Warwick, said: “The algorithm we have developed allows us to take fifty candidates across the threshold for planetary validation, and upgrade them to real planets. We hope this apply technology to large examples of candidates from current and future missions such as TESS and PLATO.
“In terms of planetary validation, no one has ever used a machine learning technique before. Machine learning has been used to rank planetary candidates, but never within a probabilistic framework, which is what you need to truly validate a planet. Instead of saying which candidates are more likely to be planets, we can now say what the exact statistical probability is. Where there is less than a 1% chance that a candidate is a false positive, it is considered a valid planet. ”
Dr. Theo Damoulas of the Department of Computer Science at the University of Warwick, and Deputy Director, Data Centric Engineering and Turing Fellow at The Alan Turing Institute, said: “Probabilistic approaches to statistical machine learning are particularly suited to an exciting problem such as this in astrophysics which requires integration of prior knowledge – from experts such as Dr. Armstrong – and quantification of uncertainty in predictions.A prime example when the extra computational complexity of probabilistic methods significantly tells.
Once built and trained, the algorithm is faster than existing techniques and can be fully automated, making it ideal for analyzing the potential thousands of planetary candidates observed in current research such as TESS. The researchers suggest that it should be one of the tools to use collectively to validate planets in the future.
Dr. Armstrong adds: “Nearly 30% of the known planets to date have been validated using only one method, and that is not ideal. Developing new methods for validation is only desirable for that reason. But machine learning also leaves us very tired. do fast and prioritize candidates much faster.
“We still have to spend time training the algorithm, but once that is done it will be much easier to adapt it to future candidates. You can also add new discoveries to gradually improve it.
“A survey like TESS provides tens of thousands of planetary candidates and it is ideal to analyze them all consistently. Fast, automated systems like these that can take us all the way to validated planets in fewer steps, let us do that efficiently. ”
Reference: “Exoplanet Validation with Machine Learning: 50 New Validated Kepler Planets” by David J Armstrong, Yevgeny Gamper, Theodoros Damoulas, August 20, 2020, Monthly Announcement from the Royal Astronomical Society.
DOI: 10.1093 / mnras / staa2498
The research of Dr. Armstrong was supported by the Science and Technology Facilities Council (STFC), part of UK Research and Innovation, through an Ernest Rutherford Fellowship.
