When: Wednesday, November 16, 2022, 3:30 PM - 5:00 PMWhere: Physics (G. O. Jones building) Room 516 & online
Speaker: Alessio Spurio Mancini (UCL)
Bayesian inference represents a rigorous approach to uncertainty quantification in cosmology. However, inferring cosmological parameters from astronomical datasets imposes large computational demands, which will become prohibitive in the analysis of data from future surveys. In my talk I will present COSMOPOWER, an open-source Python framework for Deep Learning -- accelerated Bayesian inference of cosmological parameters from next-generation Cosmic Microwave Background and Large-Scale Structure surveys. COSMOPOWER provides orders-of-magnitude acceleration to the inference pipeline by training Deep Learning emulators of matter and CMB power spectra. Bayesian parameter contours can thus be recovered in just a few seconds on a common laptop, as opposed to the many hours, days or months of runtime on computer clusters required by standard methods. This opens up the possibility of exploring unprecedentedly large parameter spaces, thoroughly accounting for systematics in the final constraints. I will conclude my talk with an application of COSMOPOWER to the study of the degeneracy between modified gravity and baryonic feedback nonlinear effects, two of the most uncertain aspects in the theoretical modelling of cosmological observables.