Satellite signal extraction with Monte Carlo Markov Chains

Our universe is filled with billions of galaxies which are detected by current research satellites. Due to gravitational attraction, these galaxies cluster such that satellite images of the night sky reveal a spidery web of galaxies all throughout our Universe. This project's aim is to improve one of the world's current best codes to extract information on gravity, dark energy, dark matter and other astrophysical processes from such maps of galaxies.

The students will tune an already existing c++ code that MCMC samples such sky data. The project has sub-components of different tuning algorithms for the sampler. Careful: This sampler runs in 17 million dimensions and solves a huge missing data problem! We therefore also encourage self-proposed ideas for how to improve the efficiency of the code, as it can easily output terrabytes of data.

Our MCMC sampler has a tuning phase. The tuning is more successful if the sampler creates more effective samples per second. Any improvement to the code that increases the effective number of samples per second are a minimum-viable fully completed version of the project.

The workflow for the students will be: 1) Understand the structure of the astronomical data: random maps on the sphere 2) Understand MCMC and why it needs tuning 3) Select from (known to us) a list of papers which improvement to the sampler tuning they'd like to implement; OR propose their own improvement 4) Demonstrate (by running the code) how the effective number of samples per second increases 5) Possibly (for personal fun of the students) make movies of which structures the sampler discovers in the web of galaxies. The picture above shows the satellite data on the sphere, and the sampler's analysis of these sky data.