‘Dark Emulator’: researchers at Kyoto University develop first AI tool for space exploration
※This is a machine-translated text of the original article.
Researchers at Yukawa Institute for Theoretical Physics of Kyoto University, Associate Professor Takahiro Nishimichi and others, announced that they had developed the AI tool “Dark Emulator,” which predicts the evolution of large-scale structures in space.
The dark simulator learns their correspondence from 101 virtual universes calculated by varying the amount and nature of the dark components of the universe.
This made it possible to make a theoretical prediction of the expected observation results for a new space model at high speed without conducting new simulations.
The virtual universe data used for this learning was calculated over about three years using supercomputers ATERUI and ATERUI II at the National Astronomical Observatory of Japan, and abundant simulation data with a total capacity of about 300 terabytes.
The tool can predict the spatial distribution of galaxies and the actual observation of the weak gravity lens effect with an accuracy of about 2-3 percent.
In addition, standard laptops could make theoretical predictions within a few seconds, reducing the computational cost to about one-hundred-millionth.
The results of this research were published on October 8, 2019 in the U.S. Astrophysical Journal, a magazine specializing in astrophysics.
Original Text: https://ampmedia.jp/2020/02/06/ai-cosmo/