Abstract: The upcoming generation of cosmological surveys such as LSST will aim to shed some much needed light on the physical nature of dark energy and dark matter by mapping the Universe in great detail and on an unprecedented scale. While this implies a great potential for discoveries, it also involves new and outstanding challenges at every step of the science analysis, from image processing to the modeling of astrophysical systematics. In this talk I will illustrate how recent advances in Deep Learning open new perspectives for addressing some of theses challenges and for exploiting this wealth of data in new and exciting ways. As a first example, I will present our work on automated strong gravitational lens detection, a problem made tractable at the scale of LSST by Deep Learning by essentially eliminating the need for human visual inspection. In a second example of applications, I will illustrate how data-driven deep generative models can be used to complement a physical modeling in two different situations: image simulations with realistic galaxy morphologies for the calibration of weak lensing shape measurement algorithms, and the production of mock galaxy catalogs with realistic intrinsic alignments learned from hydrodynamical simulations.