Transformers for Dark Matter Morphology with Strong Gravitational Lensing

Strong gravitational lensing is a phenomenon where the light of distant galaxies is bent and distorted by the gravity of massive galaxy clusters, which contain dark matter. Strong gravitational lensing has proven to be an effective method for detecting the substructures of dark matter halos. A substructure provides essential information for the identification of the true nature of dark matter. Recently, Deep Learning techniques have been quite successful in classifying different substructures and predicting the mass densities using images from the simulation of strong gravitational lensing. DeepLense is one of the projects that uses state-of-the-art Deep Learning methods to exploit this information from the underlying substructures. This project will, thus, focus on the further development of the DeepLense pipeline to integrate Vision Transformers as another approach to the existing implementations of ResNets and Equivariant Networks. We will be implementing and benchmarking various versions of Vision Transformers to achieve a robust architecture with high metrics for classification tasks respectively on the simulated strong lensing images.