Differentiable probabilistic programming for strong gravitational lensing

The difficult task of observing Dark Matter subhaloes is of paramount importance since it would constrain Dark Matter particle properties (cold or warm relic) and confirm once again the longstanding $\Lambda$CDM model. In the near future the new generation of ground and space surveys will observe thousands of strong gravitational lensing systems providing a unique probe of Dark Matter substructures. Here, we describe a new strong lensing analysis pipeline that combines deep Convolutional Neural Networks with physical models and exploits traditional sampling techniques such as Hamiltonian Monte Carlo. Using simulated strong gravitational lensing systems, we discuss first results and characterize the accuracy of the reconstruction of the main lensing parameters.