Chinese scientists develop AI model for better space observation

​​​​​​​By Saadet Gokce

ISTANBUL (AA) - Chinese scientists have developed an artificial intelligence (AI) model that improves space observation by reducing noise, according to a study published in the journal Science.

The Astronomical Spatiotemporal Enhancement and Reconstruction for Image Synthesis (ASTERIS) uses computational optics and AI algorithms, according to the state-run news agency Xinhua.

It was developed by a cross-disciplinary research team from Tsinghua University.

The model can extract extremely faint astronomical signals, identify galaxies more than 13 billion light-years away, and generate "the deepest deep-space images ever produced."

By applying the model's "self-supervised spatiotemporal denoising" technique to data from the James Webb Space Telescope, observational coverage from visible light is extended at around 500 nanometers to the mid-infrared at 5 micrometers.

"Benchmarking on mock data indicates that ASTERIS improves detection limits by 1.0 magnitude at 90% completeness and purity, while preserving the point spread function and photometric accuracy," according to the study’s abstract that was published Thursday.

With the model, the researchers were able to identify more than 160 candidate high-redshift galaxies from the "Cosmic Dawn" period, about 200 million to 500 million years after the Big Bang, triple the amount of discoveries made with previous methods, according to Cai Zheng, associate professor at Tsinghua's Department of Astronomy and a member of the research team.

The model identifies subtle noise fluctuations and differentiates them from the ultra-faint signals of distant stars and galaxies, with a "photometric adaptive screening mechanism."

- Chinese team develops rare disease diagnosis AI system with higher accuracy

Separately, another Chinese research team developed an AI system called DeepRare to help diagnose rare diseases, according to a study published Thursday in the journal Nature.

The system has set a new record for diagnostic accuracy.

The system's first-attempt accuracy was 57.18% in phenotypic diagnosis, which is nearly 24% higher than the previous global model. With genetic data incorporation, the accuracy was up to 70%.