NASA scientists used the ExoMiner deep learning neural network to distinguish real exoplanets from other signals recorded by the Kepler telescope.
NASA researchers uploaded data from the Kepler telescope in a new neural network called ExoMiner. The algorithm produced amazing results – it discovered 301 exoplanets, however, not a single of the terrestrial type. Nevertheless, in the future, the neural network will be much more actively used in the search for space objects.
Neural networks: The future of astronomy?
Why is ExoMiner so useful?
ExoMiner complements the people who professionally analyze data from telescopes to determine which signals are coming from planets and which are not. But without this neural network, solving such a problem would take a huge amount of time – thousands of stars fall into Kepler’s field of view, each of which may have several exoplanets. In addition, the ExoMiner is more accurate than other machine classifiers and people who are biased.
Transit method
Usually, planets are found by the transit method – a celestial body revolves around a star, and when it is in front of it, astronomers record a change in its brightness. The neural network is trained on past confirmed cases and false positives, which allows it to distinguish the planet from other objects and interference.
Pleiades supercomputer
Unlike other neural networks, ExoMiner is not a “black box” – that means that scientists have identified clear characteristics for exoplanets that help the AI distinguish the different objects. In its work, the neural network uses a NASA supercomputer – Pleiades, which holds 81st place in the world of supercomputers with 241,108 cores and a maximum power of 5951 teraflops.
None of the 301 planets discovered by ExoMiner are Earth-like
According to preliminary data, none of the planets confirmed with the help of ExoMiner are similar to Earth and are not located in the habitable zone of their stars. Having trained the neural network on Kepler data, scientists are going to apply it on other missions, in particular, to analyze data from the TESS space telescope, which was launched from Earth in 2018 to search for exoplanets.
How many exoplanets have we discovered?
For a long time, the problem of finding planets outside the solar system was insoluble due to the fact that these celestial bodies are too small and dim. The first exoplanets were discovered in the early 1990s. Before the use of ExoMiner, scientists had confirmed about 4,575 bodies. With the new update, their number will increase to nearly 4,900.
Besides those confirmed exoplanets, NASA has a long catalog of nearly 8000 objects that have yet to be confirmed. Most of this is the result of Kepler’s observations – the first-ever exoplanet hunter operated for nearly a decade. Even years later, scientists are a long way from analyzing its entire database and ExoMiner will have a lot more work in the future.
Not to mention that the Tess telescope has been even more effective in collecting data and ESA also plans to launch its own planet hunter in 2026 – the PLATO Telescope, which will focus on searching for exoplanets near yellow and orange dwarfs, which are similar to our Sun.
Join the discussion and participate in awesome giveaways in our mobile Telegram group. Join Curiosmos on Telegram Today. t.me/Curiosmos
Sources:
• Lewis, B. (2021, November 23). UCLA astronomers discover more than 300 possible new exoplanets. UCLA.
• NASA. (n.d.). New Deep Learning Method adds 301 planets to Kepler’s total count.
• Pultarova, T. (2021, November 28). Ai discovers over 300 unknown exoplanets in Kepler Telescope Data. Space.com.
• Williams, M. (2021, November 27). A machine-learning algorithm just found 301 additional planets in Kepler Data. Universe Today.
