The reporter learned from the National Astronomical Observatory of the Chinese Academy of Sciences on the 25th that scientific research teams from the National Astronomical Observatory of the Chinese Academy of Sciences, the University of Chinese Academy of Sciences and other units have successfully developed an artificial intelligence model called SpecCLIP. This model is like a translator who is proficient in multiple "stellar languages" and uniformly interprets stellar spectral data from different telescopes, providing a new technical tool for astronomical research. Relevant research results were published in The Astrophysical Journal.
Stellar spectra are often compared to scientists studying the universe's "fingerprints." Each star's spectrum hides its identity—its temperature, chemical composition, and surface gravity. By analyzing these "chemical signatures", astronomers can trace the evolution of the Milky Way from its birth to the present, just like archaeologists reconstructing history.
However, real-life research faces a big problem: different sky survey projects, such as China's Guo Shoujing Telescope (LAMOST) and Europe's Gaia satellite, obtain spectral data in different ways, resolutions and band ranges. These data are like stories told in different dialects and are difficult to put together directly for large-scale analysis.
The SpecCLIP model was born precisely to break this data barrier. The research team innovatively introduced ideas similar to the "big language model" into the field of astronomy, using the "contrastive learning" method to allow AI to automatically learn and establish internal connections between spectral data from different sources.
SpecCLIP can not only predict the atmospheric parameters and element content of stars at once, but also conduct spectral similarity searches and even help discover special celestial objects. Based on its powerful data unified representation capabilities, SpecCLIP has played a role in many cutting-edge scientific explorations. For example, in the "Earth 2.0 (ET)" mission to search for the second Earth, it can accurately characterize the characteristics of the planet's host star, thereby improving the screening efficiency of potentially habitable planets.
