This repository contains the implementation of FusionNet designed to jointly perform graph and node classification for particle reconstruction in HEP. By learning optimized representations for both nodes and the overall graph structure, FusionNet improves property estimation for each task.
This model has been applied in CMS for jointly identifying b-jets and tracks from b-hadron decays where it outperforms existing methods in the track identification problem.
Requirements: pytorch, pytorch-geometric
Developers: Prince Kumar (kumarprinceraj2004@gmail.com), Shamik Ghosh (sameasy2006@gmail.com)