Researchers: Thomas Gaskin, Marc Timme (TU Dresden), Mark Girolami
This project aims to analyse and generate large graphs with specified topological properties using a Kronecker decomposition. This then allows using e.g. gradient descent methods to sample the Pareto front given by the constraints, and find suitable Kronecker factors that, when combined, produce a larger graph with the desired properties. This method makes use of the efficiency of the Kronecker product, allowing for fast generation of large graphs, while avoiding the low variance in graph structure previous methods have suffered from without resorting to stochasticity.