New AI tool can eliminate


Japanese astronomers have developed a new artificial intelligence (AI) technique that can eliminate "noise" in astronomical data due to random variations in galaxy shapes, obtaining results consistent with currently accepted models of the universe, according to a recent report. This powerful new tool can be used to analyze big data from current and planned astronomical surveys, according to the researchers.


The researchers explain that scientists can use gravitational lensing techniques to study the large-scale structure of the universe, but in gravitational lensing, foreground objects (such as galaxy clusters) distort images of background objects (such as distant galaxies), and some galaxies are inherently odd-looking, so this gravitational lensing technique, which looks at images of many galaxies, runs into a problem: it is difficult to distinguish whether the distortions in their images are caused by gravitational This is called shape noise, and is one of the limiting factors in studying the large-scale structure of the universe.


To eliminate shape noise, a group of Japanese astronomers first used the world's most powerful astronomical supercomputer, the Atreui II, to generate a catalog of 25,000 simulated galaxies based on real data from the Subaru telescope. They then added real noise to these perfectly known artificial datasets and trained the AI to recover foreground material from the simulated data.

After training the AI tool to recover fine details previously unobservable, the team then used this tool to process real data covering 21 square degrees of sky and found that the distribution of foreground matter masses was consistent with the standard cosmological model.

The team explained, "This study shows the benefits of combining different types of studies such as observations, simulations and artificial intelligence data analysis. In this era of big data, we need to cross the traditional boundaries between specialties and use all the tools available to make sense of the data. If we can do this, it will

  • TAGS