In the field of astrophysics, two primary observational techniques shape our understanding of the cosmos: photometric and spectroscopic measurements. Photometric imaging captures the total light emitted by celestial objects, enabling comprehensive all-sky surveys that catalogue vast numbers of galaxies. Spectroscopic analysis, more intricate and time-intensive, splits light into a detailed "rainbow" spectrum, revealing critical insights about a galaxy's composition, velocity, and stellar makeup. While photometric methods provide a broad overview, spectroscopic measurements offer a deeper, more nuanced understanding, but for a smaller subset of observed galaxies.
Eva Sextl and Lars Doorenbos, a computer scientist at the University of Bern, had the idea of using generative artificial intelligence (GenAI) to predict detailed optical spectra of galaxies from their images in different colours. Although a spectrum contains much more information than a simple image, it works: In a pilot study, Eva Sextl and Lars Doorenbos trained their algorithm with images and spectra from the Sloan Digital Sky Survey (SDSS). They were able to show that the information content of the artificially generated spectra is basically identical to that of the real spectra in the test set.
Next, the two will train their GenAI algorithm with the cosmological data sets of the Dark Energy Spectroscopic Instrument (DESI) to analyse galaxies at higher redshift. They want to find out whether their method can detect a distant class of galaxies with pronounced spectral lines, so-called Lyman-alpha emitters (LAE). "This research stands out for its bold integration of cutting-edge generative AI with astrophysical analysis with the potential to redefine how we study the universe, showcasing scientific vision and a deep interdisciplinary understanding," said the jury.
The AI Hub Prize is sponsored by the Munich University Society and supports the project costs of young researchers.
Press release about the opening of the AI-Hub@LMU
Publication:
L. Doorenbos, E. Sextl, K. Heng, et al., „Galaxy Spectroscopy without Spectra: Galaxy Properties from Photometric Images with Conditional Diffusion Models“ ApJ 2024
Contact:
Eva Sextl
Ludwig-Maximilians-Universität / Excellence Cluster ORIGINS
Email: sextl(at)usm.lmu.de
