A group of scientists led by the Leibniz Institute for Astrophysics Potsdam (AIP) and the Institute of Cosmos Sciences at the University of Barcelona (ICCUB) have developed a novel machine learning model that efficiently processes data for 217 million stars observed by the Gaia mission.
The results rival traditional methods for estimating stellar parameters, opening new avenues for mapping characteristics such as interstellar extinction and metallicity across the Milky Way, which enhances our understanding of stellar populations and the galaxy's structure.
With the European Space Agency's Gaia mission releasing its third dataset, astronomers now have access to improved measurements for 1.8 billion stars, presenting a significant challenge in data analysis. The researchers employed machine learning to estimate crucial stellar properties using Gaia's spectrophotometric data. Their model, trained on high-quality data from 8 million stars, achieved reliable predictions with minimal uncertainties.
First author Arman Khalatyan from AIP explains, 'The technique, called extreme gradient-boosted trees, allows for precise estimation of stellar properties such as temperature and chemical composition with unprecedented efficiency. The SHBoost model can complete its tasks in four hours on a single GPU, a process that previously took two weeks using 3,000 high-performance processors.'
This innovative approach significantly reduces computational time, energy consumption, and CO emissions. It is the first successful application of this technique to stars of all types simultaneously.
The model utilizes high-quality spectroscopic data from smaller surveys and applies this knowledge to Gaia's extensive dataset, extracting key stellar parameters using only photometric and astrometric data, along with low-resolution XP spectra.
According to Cristina Chiappini from AIP, 'The high quality of the results decreases the need for additional resource-intensive spectroscopic observations when identifying candidates for further studies, such as rare metal-poor or super-metal-rich stars, which are vital for understanding the early phases of Milky Way formation.'
This technique is crucial for preparing future observations with multi-object spectroscopy, including the 4MIDABLE-LR survey, part of the 4MOST project at the European Southern Observatory (ESO) in Chile.
Friedrich Anders from ICCUB adds, 'The new model provides extensive maps of the Milky Way's chemical composition, confirming the distribution of young and old stars, and revealing concentrations of metal-rich stars in the galaxy's inner regions.'
The team also mapped young, massive hot stars throughout the galaxy, highlighting poorly-studied regions where star formation occurs. The data indicate the presence of 'stellar voids' in the Milky Way—areas with very few young stars—and show where the three-dimensional distribution of interstellar dust remains poorly understood.
As Gaia continues its data collection, the ability of machine learning models to handle vast datasets quickly and sustainably positions them as essential tools for future astronomical research, demonstrating the potential to revolutionize big data analysis in astronomy and promote more sustainable research practices.