Time delay estimation in unresolved lensed quasars

The fractional uncertainty on the $H_0$ measurement, performed with time-delay cosmography, linearly decreases with the number of analysed systems and it is directly related to the uncertainty on relative time delays measurements between multiple images of gravitationally-lensed sources. Analysing more lensed systems, and collecting data in regular and long-term monitoring campaigns contributes to mitigating such uncertainties. The ideal instruments would clearly be big telescopes thanks to their high angular resolution, but, because of the very large amount of observational requests they have to fulfill, they are hardly suitable for the purpose. On the other hand, small/medium-sized telescopes are much more accessible and are often characterized by more versatile observational programs. However, their limited resolution capabilities and their often not privileged geographical locations may prevent them from providing well-separated images of lensed sources. Future campaigns plan to discover a huge number of lensed quasar systems which small/medium-sized telescopes will not be able to properly resolve. This work presents a deep learning-based approach, which exploits the capabilities of convolutional neural networks to estimate the time-delay in unresolved lensed quasar systems. Experiments on simulated unresolved light curves show the potential of the proposed method and pave the way for future applications in time-delay cosmography.