Creating an artificial intelligence tool that allows high-resolution data capture on exoplanets, contributing to the European Space Agency’s mission, has led researcher Louis Simois to win the Ariel Machine Learning Data Challenge.
The award came after Louis Simoes improved his ability to detect exoplanets, by capturing the light emitted by a star, using a high-resolution algorithm.
The challenge was launched by the European Space Agency (ESA) to ensure the necessary conditions for studying exoplanets (planets that orbit a star, but do not belong to the solar system) on the Ariel mission, which will be launched in 2029.
For the Portuguese researcher, who “periodically participates in ‘space’ machine learning competitions”, this competition had the advantage of developing his “ability to tackle problems and apply various algorithms”, as well as keeping up with “developments in artificial intelligence”.
For the challenge, which he thought was “excellent, given its relevance and complexity,” he responded by creating a solution with a mean margin of error of 0.00007.
“The specific problem, which is the discovery of exoplanets, is done through different approaches, but the most successful one is to study the curves of light and, therefore, the light that comes to us from a star and when the planet, in its orbit, passes forward, the shape that this curve assumes tells us a lot about the planet ‘,” Luis Simويسes explained to Lusa.
The method has already been used to “identify thousands of exoplanets,” he acknowledges, “but this mission has a very ambitious goal,” which includes “understanding the chemical composition of these planets’ atmospheres.”
“This would only be possible with tools that do not exist today, hence the interest, 10 years ago, to start the study and launch the challenge to the community,” added Louis Simois.
His solution does not provide new data, but Louis Simoes created “a model that has practically no errors (…), a very accurate model”.
This accuracy is important, because “the question of deducing the chemical composition of the atmosphere is a later processing of those data that the algorithm will give.”
“There, it is already entering other areas of astrophysics, but the point is that if a wrong interpretation is made at this step in the data interpretation chain, the next steps will be misled about the true characteristics of the planet,” he said.
To get there, this algorithm was programmed “automatically, based on training data”.
For this, the team leading the ESA mission “generated synthetic data, through simulators, which, as faithfully as possible at the moment, recreate the future data that the mission will collect.”
These data were corrupted “because of the kinds of data corruptions that will appear in the mission – because of thermal fluctuations, because of the whole challenge of measuring hundreds of light-years how many photons are coming from different sources”, but knowing what the raw data was, it was possible to assess the level of accuracy Offered solutions.
“We’re still years away, and that’s not the final model yet, but it’s already a big step toward achieving the levels of accuracy that the mission would like to have,” the researcher said.
Luís Simões began work on applying artificial intelligence to space problems, having produced “Airbus and ESA systems for controlling ship landings on other planets”.
In 2008, with the crisis, he moved to the Netherlands, where he began cooperation with the European Space Agency.
He returned to Portugal in 2018, where he and his wife created ML Analytics.
“Friendly zombie fanatic. Analyst. Coffee buff. Professional music specialist. Communicator.”
