Cosmology and beyond: Solutions for high-dimensional parameter estimation
Postgraduate Student, Institute for Astronomy, University of Edinburgh
Abstract: Cosmology, from the beginning of this century onward, made heavy use of Bayesian statistics to constrain fundamental parameters of the universe. These include, for example, important insights into dark matter and dark energy to better understand our universe. Current efforts in cosmological parameter estimation, however, often suffer from the computational costs of approximating high-dimensional distributions with expensive likelihood calculations. This is not a problem exclusive to cosmology, but also in areas as diverse as reliability engineering, computational linguistics, econometrics and finance, and genetic analysis. In a time in which clustered computational resources are comparably cheap and time is precious, general-purpose approaches that extend previous methods are needed to solve this problem, preferably in the form of easy-to-handle and open-source software. This talk covers the cosmological background and rationale that inspired the developed methodology, as well as an explanation of an approach that makes use of recent advances in Bayesian statistics, sampling methods and parallel computing. In the background section, participants will learn about modern cosmology, the role statistics plays in it, and planned telescopes on the ground and in space. In the methodological part, the talk will provide a better understanding of bottlenecks and ways to solve them in parameter estimation problems. In addition, a Python package implementing this project is presented to make current research accessible to the wider community.
Bio: Ben is a PhD student at the University of Edinburgh's Institute for Astronomy, where his primary research covers Bayesian nonparametrics for parallelized high-dimensional parameter estimation, and the merging of analytic models of galaxy evolution into deep learning frameworks. He previously obtained a master's degree in AI and worked as a machine learning scientist in the UK's investment management industry.