When Less is More: Lightweight and Distilled GNNs for Efficient Particle Reconstruction

Poster presenting strategies to reduce the structural complexity of GNNs for real-time LHCb calorimeter reconstruction: omitting explicit edge features, replacing full message passing with learnable permutation-invariant aggregation (up to 8× faster inference), and knowledge distillation of compact student models.

Sep 1, 2025 — Sep 5, 2025