Speaker: Adam Rouhiainen, Physics PhD Graduate Student
Abstract: The large-scale structure is highly non-Gaussian at late times and small length scales, making it difficult to describe analytically. Parameter inference, data reconstruction, and data generation are greatly aided by various machine learning models, and this work takes a field level approach to solving these problems. The probability distribution of the large-scale structure is learned with normalizing flows, allowing Bayesian reconstruction of noisy fields with removed foregrounds. The normalizing flow is trained to be conditional on cosmological variables, from which accurate parameter estimation can be done. Turning to highly expressive denoising diffusion models, a super-resolution emulator is developed for large cosmological simulation volumes, allowing high-resolution simulation volumes to be conditionally generated from low-resolution volumes. The super-resolution emulator is trained to perform outpainting, and can thus upgrade very large cosmological volumes from low-resolution to high-resolution using an iterative outpainting procedure.