Classical void finders depend on hand-tuned choices: smoothing scales, density thresholds and merging criteria. In this project I treat void finding as a node-classification problem on the graph given by the Voronoi tessellation of the tracers, and train a deep graph-attention network to reproduce a void catalogue directly from it, in an N-body simulation. It is a side project, in which I also build my skills in machine learning.
Side project · CPPM · ongoing
The full description is behind a password until publication. Drop me an email if you'd like access.