In the near future, upcoming Large-Scale Structure (LSS) missions, including DESI, eROSITA, Euclid, WFIRST, and LSST, are poised to survey extensive cosmological volumes, collecting terabytes of data that hold the promise of revolutionizing our understanding of cosmological parameters with unparalleled precision. To attain this ambitious goal, it is imperative to maximize the information extracted from this data. Nevertheless, we face two pivotal challenges in achieving optimal cosmological analyses: the development of precise theoretical models within the non-linear regime and the creation of innovative computational techniques to overcome the computational bottlenecks inherent in conventional simulation methods. During this presentation, I will introduce a cutting-edge machine learning (ML) approach tailored to the construction of a deep learning emulator at the field level for cosmological simulations. I will illustrate how the deep learning methodology provides a highly accurate alternative to conventional techniques by directly translating non-standard cosmological simulations, including those involving neutrinos, from standard simulations. This approach holds the potential to generate precise predictions for cosmological fields across a spectrum of input parameters, streamlining the exploration of non standard cosmological scenarios while significantly enhancing efficiency.

In the latter part of the talk, I will also discuss the exciting potential of utilizing deep learning-based approaches to model the connection between the underlying dark matter distributions from N-body simulations and the galaxy distributions derived from full hydrodynamic simulations. This multifaceted approach promises to significantly advance our capabilities in understanding and modeling complex cosmological phenomena.