Rapid Cosmology: machine learning emulators and likelihood analysis

Project Description

Cosmology is a new and fast growing field, owing this growth to both new technology and statistical methods in imaging and data processing. A successful cosmological study relies on analysing a large amount of data against cosmological models with varying parameters. One of the biggest questions in cosmology is: “what is the nature of dark matter and dark energy?” which my research aims to answer.

The modelling of cosmological observables can be done with theoretical pipelines in some cases, however many still rely on simulations. Both methods can be slow, although simulations tend to be much slower, given that they model the whole Universe! The recent advancements in machine learning and artificial intelligence have opened an opportunity for substantially increasing the speed of analysis, through the use of emulators. These emulators can mimic the features in the simulations or theoretical pipelines and effectively replace them.

I will develop, train and utilise emulators for both theoretical and simulation based pipelines. I will use them to analyse data from the state-of-the-art Kilo Degree Survey (KiDS), where I lead the modelling effort for our final data release. With the substantially increased speed I will be able to shed light on the nature of dark matter and dark energy.

This work will form a basis for future cosmological analysis using data from large scale surveys such as Euclid and LSST (Rubin Observatory). This work is also very timely as it will substantially decrease the environmental impact of running cosmological studies using supercomputers.