Scientists validated 50 new planets with AI — if you needed inspiration

It seems the power of machine learning is limited only by available data and imagination.
28 August 2020

A starfield (elements of this image furnished by NASA). Source: Shutterstock

  • A new machine-learning algorithm designed by astronomers and computer scientists from the University of Warwick confirms new exoplanets in telescope data
  • Sky surveys find thousands of planet candidates, and astronomers have to separate the true planets from fake ones

As humans, with jobs, pressing tasks, and, well, lives to live, we don’t often have the time or expertise to manually crawl through masses of data in the quest for new insights and intelligence.

Instead, we are increasingly applying artificial intelligence (AI) and machine learning techniques to do this for us. It can do it in a fraction of the time, and spot patterns that we might miss. The business benefits of this potential are obvious. An AI program could be let loose into vast reserves of dark data, turning out valuable, hidden customer insights, or enabling marketplace predictions. In industries like healthcare, this approach is serving as a ground-breaking new method of diagnosis, able to filter through tens of thousands of MRI scans, x-rays, and other medical records, identifying connecting patterns and correlations between patients and cases — all in comparative lightspeed compared to a medical practitioner.

The applications are only limited by imagination and available data, and there are perhaps few so powerful examples of AI’s utility in this respect than it being deployed to actually discover new planets in the vast expanse of space. That’s exactly what scientists from Warwick University have done — and if you needed some inspirational motivation to employ machine learning techniques, this might well be it. 

Researchers used a machine-learning algorithm to confirm the existence of 50 new planets. In much the same way as described above, the scientists fed the algorithm huge datasets from NASA’s retired Kepler mission and Transiting Exoplanet Survey Satellite (TESS) space telescope launched in 2018. 

In what’s hoped to be a new approach for future planet validation, the algorithm was trained to differentiate between planets and false positives amid the noise of and inference caused by objects in the background and occasional errors in the camera. It was then launched into a separate, unvalidated data set on a mission to validate celestial bodies. 

Dr. David Armstrong, from the department of physics at the University of Warwick, said: “The algorithm we have developed lets us take fifty candidates across the threshold for planet validation, upgrading them to real planets.” The program also generated the probability of each planet being real. Those planets validated ranged from the size of Earth to the size of Neptune, with orbits varying from one day to 200 days. 

“A survey like TESS is predicted to have tens of thousands of planetary candidates and it is ideal to be able to analyze them all consistently,” said Armstrong.

“Where typical sky surveys identify thousands of potential planets, it’s left to human astronomers to distinguish real planets from the information. Fast, automated systems like this that can take us all the way to validated planets in fewer steps let us do that efficiently.

“We still have to spend time training the algorithm, but once that is done it becomes much easier to apply it to future candidates.”

TechHQ previously explored how AI could enhance discoveries in the field of materials science. Lawrence Berkeley National Laboratory used machine learning methods to capture latent knowledge from approximately 3.3 million abstracts, primarily focused on materials science, physics, and chemistry. 

By using nothing more than patterns in text, an algorithm made valid scientific predictions.

Dr. Anubhav Jain, one of the study’s co-authors, previously told TechHQ: “You could imagine that if you did pair this technology with some of the automated laboratories, you could have a system in the future where AI reads the research papers, figures out which experiments to do, actually conducts those experiments, and maybe even pick a materials formulation that is useful for some purpose, all on its own. 

“I think it’s theoretically possible to do something like that but there’s still a lot of practical issues in all of these things that prevent it from being right around the corner.”