How scientists are using artificial intelligence to find extraterrestrial Earths

Machine learning algorithms trained in synthetic planetary systems were released to identify almost four dozen real stars that are likely to host rocky planets in habitable zones.

"This model has identified systems that are highly likely to accommodate planets like Earth that have not been discovered," says Jeanne Davy, an astronomer at the German aerospace agency DLR. "Another study has confirmed the theoretical opportunity for these systems to organize planets like Earth."

Often, in the sense that they have planet-like masses and live in habitable zones of stars, the "Earth-like" world Earth is done in large-scale investigations where thousands of stars are often observed to pass through planets. However, astronomers also want to find the possibility of finding planets in habitable zones the size of Earth, so they need a more targeted measure to find candidate stars.

He was the one who Davoult encouraged the develop the algorithm while she was in Switzerland at the University of Bern. He had to train his data to learn to identify all models, patterns based on machine learning algorithms and make predictions based on algorithms that display these patterns.

The problem, however, is that 6,000 exoplanets have been discovered so far, stained with information about these worlds. Generally, even in a world of 6,000 is not enough to train an algorithm.

Davort and her colleagues, Roman Eltsinger and Jan Alibert, of Baan University, turned to another model that can simulate the world based on everything they know about the planetary system.

The Bern model for planet formation and development has been continuously implemented at the University of Bern since 2003, and is constantly improving as more data and theoretical models are available.

"We can use the Bern model to create a statement on how planets develop under certain conditions on protoplanet-shaped hard drives, under certain conditions, how they are developed, and which planets develop," Alibert said in the description.

"The Bern model is one of the few models worldwide that provides such interconnected physical processes and allows for research like current implementations."

The Bern model simulated 53,882 simulated planetary systems by three different types of stars: G-shaped stars such as our Sun, red dwarves about half the Sun, and a group of second-generation red dwarves with only one-fifth of the Sun's mass.

An algorithm that searches these simulated planetary systems for patterns or correlations. Thus, the presence or absence of habitable planets of the size of Earth is combined with the various architectures of the planetary system.

Some correlations are more obvious than others. For example, there is a correlation between the presence of an inner rocky planet that listens to a system with an outer gas giant. This is the same architecture as our solar system, with rocky planets approaching the sun more than the gas giants. On the

flip side, there was an anticorrelation between the gas giant Hot Jupiter near the sun and the "pea" planet, a rocky planet of similar mass and orbital distances, arranged by stars of red stars such as Trapist-1 and Barnard's Stars.

Hot Jupiter is a gas giant and continued to form from its stars before wandering and knocking along its way, so we didn't expect to find Hot Jupiter next to such an ordered rocky planet.

However, there is also a deeper correlation identified by Davoult in previous studies. In particular, the mass, radius, and orbital period of the innermost verifiable planet seem to be the main guide, whether the system houses moderate planets of Earth size. For example,

found that the presence of planets within habitable zones of Earth's size by G-shaped, like our Sun, is more likely if the innermost detectable planet has a larger radius than Earth, or if there is an orbital for more than 10 days. With knowledge of these correlations, the algorithm was successfully trained with simulated data.

"The results are impressive. The algorithm reaches a precision value of up to 0.99. That is, 99% of systems identified by machine learning models have at least one Earth-like planet," Davoult said. Confidently, in the ability of the algorithm to recognize correlations,

Candidates from the planetary system were then hired for actual observations, if a planet the size of the Earth is likely to be present in the habitable zone of its planet. Astronomers can now track these goals rather than appearing blind.

The algorithm truly proves its value in the future. The European Space Agency's Platon mission is expected to discover thousands of transport planets.

By using Plato's algorithms for discovery, we should be able to narrow down thousands of systems to the few that are likely to support planets like Earth, so that astronomers can find them faster and more efficiently.