AI4Astro: Exploring Star Formation and ISM through Artificial Intelligence - Duo Xu (UVA)

Origins Seminar presented 6 November 2023
Abstract:
Machine learning, particularly deep learning, is transforming astronomy by enabling efficient processing of large datasets. Deep learning surpasses human capabilities in rapidly and accurately analyzing complex data such as images and data cubes in the field of the interstellar medium and star formation. Denoising diffusion probabilistic models (DDPMs) are machine learning algorithms inspired by diffusion thermodynamics, demonstrating state-of-the-art performance in various domains. DDPMs offer several advantages as tools for inferring physical quantities in astronomy, including stable training, robust performance, interpretability, and alignment with the inherent nature of scientific problems. In this talk, I will introduce applications of DDPMs to infer intrinsic physical quantities, such as volume density and interstellar radiation field, from observational data. I will also introduce how DDPMs can be used for segmentation tasks, such as segmenting filaments from dust emission.
Find more information about the Origins Seminar series here: eos-nexus.org/o...