Tractable generative modelling of Cosmological Structure Formation
The purpose of this project is to show that machine learning can efficiently model non-linear large scale structure formation in the universe. In particular, we use a generative model, Sum Product Networks, to predict the properties of a galaxy, given a partial history of its dynamical evolution.
The dataset is generated through the Eagle Simulations. Sum-Product Networks are the generative model used to learn cosmological structure formation.
For further introduction to SPNs, see this curated list of resources dedicated to SPNs. The SPNs used are based on the brilliant SPNFlow library.
For an introduction to Eagle simulations, see this. To use the public dataset, first register to log in to the Eagle simulations and use any of the queries provided in the folder sql_queries. The dataset can be quite huge but connection times out after 30 minutes.
