Digital Copernicus: An AI From Russia’s Urals Rewrites the Map of the Galaxy
Throughout the history of astronomy, certain discoveries have fundamentally reshaped humanity’s view of the universe. That happened when Galileo turned his telescope toward Jupiter, when Kepler formulated the laws of planetary motion, and when early instruments first detected radio signals from distant stars. Today, the galaxy’s most tireless explorer is no longer a human observer but an algorithm developed in research laboratories in Russia’s Ural region.
Secrets of Distant Stars
Scientists from the Kourovka Astronomical Observatory at Ural Federal University, working as part of a research team at NASA’s Ames Research Center, have developed a neural network designed to sift through vast volumes of astronomical measurements and identify genuine planets hidden within faint, barely perceptible changes in a star’s brightness. The algorithm, named ExoMiner++, was trained on data from the Kepler mission and the TESS mission, where most objects have already been verified and reliably classified. These NASA space telescopes search for exoplanets – planets outside our solar system – using the transit method, which detects tiny dips in a star’s brightness when a planet passes between the star and the telescope.
After analyzing 147,000 transit-like events, the neural network identified 7,330 previously unknown exoplanets. The result is striking not just in scale but in its implications – it significantly alters estimates of how densely populated the universe may be. The model also re-evaluated existing records. Of 2,506 candidates once considered promising, only 1,797 were confirmed as genuine planets.
Technology as a Cosmic Detective
The first exoplanets were discovered only in the late 20th century, and finding them required extreme precision and patience. The launch of space telescopes such as Kepler and TESS marked a turning point, flooding researchers with terabytes of data – light curves from millions of stars where a minute, periodic dip could signal a planet’s presence. Manually analyzing such volumes became physically impossible, creating a clear role for artificial intelligence.
An early breakthrough came from a neural network developed by Google, which uncovered two previously missed planets in Kepler’s archives. Later algorithms learned to filter out stellar noise and instrument interference, isolating faint signals. Each project added dozens or hundreds of new worlds to planetary catalogs. The Ural team’s result – 7,330 new planets – represents a qualitative leap. It suggests a fundamentally new model architecture or training approach capable of detecting even the smallest and most distant planets with unprecedented efficiency.
A New Statistical Picture of the Universe
The core objective of exoplanet research is to understand how unique our solar system – and Earth in particular – really is. NASA reports that the number of confirmed exoplanets has surpassed 6,000, with more than 8,000 additional candidates awaiting confirmation or rejection. Future space observatories, including the James Webb Space Telescope and the planned PLATO mission, will no longer be searching blindly. They will rely on an exceptionally large, curated list of high-priority targets.
Researchers plan to apply ExoMiner++ to new TESS datasets and adapt it for upcoming missions. The neural network is poised to become a primary tool in the search for worlds orbiting distant stars. An algorithm capable of identifying thousands of planets offers a glimpse of science’s future – one where humans define the questions and artificial intelligence navigates the labyrinth of vast data to uncover answers. The map of the galaxy has just become far denser and, paradoxically, even more expansive. Humanity now has 7,330 new destinations for its curiosity and imagination.
