When Less is More: Towards Lightweight and Distilled Graph Neural Networks for Efficient Particle Reconstruction in LHCb's Next-Generation Calorimeter

Attention-enhanced, node-centric GarNet graph neural networks achieve 8× faster inference than message-passing baselines for particle reconstruction in LHCb’s next-generation PicoCal calorimeter. A GarNet student with ~40% fewer parameters and a Graph(GarNet)→MLP distillation deliver a further 2–6× speedup at ~95% smaller model size, while surpassing the teacher on energy resolution. Firmware and trigger integration are ongoing for LHCb Upgrade II.

May 25, 2026 — May 29, 2026