MSc Project: Inter-galactic machine vision

This is a tentative proposal, as presented in 2021/22. The motivations and idea remain relevant, but the objectives have to be revised following recent developments. Please contact me to discuss options.

Purpose

The universe is full of matter that we cannot see directly, the so-called dark matter. However, dark matter can be observed indirectly because it deflects light from distant galaxies, resulting in distorted images. Good models for the distortion exists, so that we can simulate the distorted image given an undistorted image and the mass distribution of the dark matter. However, inverting this model to obtain a description of the dark matter given a distorted image is hard. The purpose of this project is to solve this inverse problem by means of machine learning.

A simulator for gravitational lensing has been developed at the department in 2022, and we are able to generate large synthetic datasets, which can be used to train machine learning models to reconstruct the lens mass. However, there are many open problems.

We propose two research questions which are suitable for an MSc thesis. A project can focus on either one or both, and there are variations available.

• RQ1: To what extent can the lens mass distribution be reconstructed from distorted images using machine learning?
• RQ2: How can the simulator be extended to simulate more generic lens models?

Contributions

The candidate will work together with astro physicists to develop tools to simulate and visualise known theories of phsyics.

Research method

Different research methods may be used depending on the angle taken. Mathematical modelling is deductive. Machine learning is evaluated with hypothetic-deductive methods, using either synthetic or empirical data. User-centric tool development is also possible, and may take more of a design approach.

Participants

• Hans Georg Schaathun (computing)
• Ben David Normann (astrofysikk)

Final deliverables and dissemination

• A working prototype available as open source. The candidate retains copyright and chooses the license. The only requirement is that the source code be available for further research.
• The dissertation which documents the scientific insights gained with respect to the learning outcomes in the module description. Good results are likely to be publishable internationally.

References

• Final year dissertation by Ingebrigtsen, Remøy, Westbø, and Austnes 2022.
• Normann, Ben David; Clarkson, Chris S.. (2020) Recursion relations for gravitational lensing. General Relativity and Gravitation. vol. 52.
• Chris Clarkson: «Roulettes: A weak lensing formalism for strong lensing - I. Overview« arXiv
• Existing simulator on github
• On methodology
  • Donald Schön (1983) The Reflective Practitioner
  • Herbert Simon (1996) Sciences of the Artificial (3rd edition)