Oral Presentation

Multi-Probe Simulation-Based Large-Scale Structure Cosmology with non-Gaussian Statistics

Author(s): Jozef Bucko (ETH Zurich)

Presenter: Jozef Bucko (ETH Zurich)

Progress in computing and machine learning has enabled an efficient extraction of information from large-scale structure beyond the Gaussian regime. In our work, we investigate the potential of combining non-Gaussian information from weak lensing and galaxy clustering fields to improve constraints on cosmological parameters. We develop a forward model based on the CosmoGrid simulation suite, allowing us to generate up to 1'000'000 independent simulated survey maps, and use neural networks to compress the data and build the likelihood, allowing for efficient SBI in 10 dimensional parameter space. We combine lensing and clustering probes at the level of peak statistics for a stage III-like survey, and showcase the gain of reaching beyond 2-point statistics.