Domain Adaptation for Decoding Dark Matter with Strong Gravitational Lensing

Dark matter is one of the biggest questions in current cosmology, and many different theories were created to try to explain it. One of the challenges of studying dark matter is actually finding it, as it is only noticeable due to gravitational interactions. Fortunately, recent results have shown that strong gravitational lensing can be used as a probe for studying dark matter’s substructure. Unsupervised learning algorithms are particularly interesting in this field because they allow for the identification of dark matter substructure without a prior theoretical model assumption. While the performance of these algorithms is very promising, there is still a large gap when compared to supervised learning algorithms. One promising possibility is to use domain adaptation techniques to fine tune the models trained on simulation data with real data. Thus, this project will focus on using domain adaptation to account for the differences in the modelling and available real observation data, while also improving the interface with PyAutoLens, the software used for creating the strong lens simulations.