Campus, Varian 255
Simulation-based inference (SBI) is a promising approach to leverage cosmological simulations and extract information from the non-Gaussian, non-linear scales that cannot be modeled analytically. However, scaling SBI to the volumes and resolutions probed by the next generation of galaxy clustering surveys can be computationally prohibitive. This is exacerbated by the fact that if we do not use accurate high fidelity simulations, SBI is susceptible to model misspecification. I will begin by putting this in context with discussing the sensitivity of SBI on the various components of cosmological simulations: gravity model, halo-finder and the galaxy-halo distribution models (halo-occupation distribution, HOD). Then, to overcome this computational bottleneck, I will present a new framework for cosmological analysis called Hybrid simulation-based inference (HySBI). HySBI combines perturbative methods (PT) on large scales with conditional SBI on small scales, thus learning the small-scale likelihood for a wide range of statistics using only small-volume simulations and drastically reduces computational costs. As a proof-of-principle, I will show results of using HySBI to constrain cosmological parameters on dark matter density fields using both the power spectrum and wavelet coefficients, finding promising results that significantly outperform classical PT methods. Finally, I will discuss a roadmap for the next steps necessary to implement HySBI on actual survey data, including consideration of bias, systematics, and customized simulations.