Letting Big Astrophysical Data Teach Us -- Unsupervised Deep Learning, Domain Adaptation

Machine learning has been widely applied to clearly defined problems of astronomy and astrophysics. However, deep learning and its conceptual differences to classical machine learning have been largely overlooked in these fields.

The general idea behind this project is to let the abundant real astrophysical data speak for itself, with minimal supervision, to detect and study the revealed patterns, which may potentially have correlations with current physical understanding of the universe, and even facilitate discovery of novel physical relationships.

The project mainly involves unsupervised representation learning based on deep convolutional networks. Depending on the pace of the project, it may also involve domain adapation techniques. You will have the opportunity to learn deep learning techniques used in the field of computer vision, and apply them on data-driven applications in astronomy and cosmology.

Current and ongoing works in the context of this project can be found here: https://www.eso.org/~nsedagha/universe/

Familiarity with Machine Learning and programming experience are required.