Arne Thomsen

Research area

I am a PhD student in Prof. Alexandre Refregier’s Cosmology Research Group and an active member of the Dark Energy Surveycall_made. The focus of my research lies at the intersection of physical cosmology and machine learning.

More specifically, I am working on simulation based inference and how the expressive power of deep learning can be leveraged in that context to extract more information from cosmological observations than traditional analysis methods permit. This overarching theme connects projects involving various data modalities, including images on the celestial sphere and point clouds, as well as different cosmological probes such as weak gravitational lensing and galaxy clustering.

My work covers the whole inference pipeline end-to-end, from processing the large numerical simulation suites serving as the theory predictions to deriving posterior constraints on the parameters. For that latter step and more broadly, I am interested in employing generative models, such as normalizing flows, in Bayesian statistics - for example, to learn conditional probability distributions from samples.

Furthermore, to be able to run all of the above, I am continously working on the computational aspects of deep learning at scale on the high performance computing clusters at CSCScall_made and NERSCcall_made.