Machine Learning for the pfRICH Particle Identification subsystem

Stony Brook University (CFNS)4th AI4EIC Workshop Machine Learning for the pfRICH Particle Identification subsystem D.H Dongwi, C-J. Naïm and L. Rhode Artificial Intelligence for the Electron Ion Collider (2025) October, 28, 2025 Stony Brook University (CFNS)4th AI4EIC Workshop 1 •A compact central detector with several subsystems" •Hermetic coverage: –3.5 < η < 3.5 (tracking, calorimetry, particle identification) p/A e pfRICH subsystem in backward region The ePIC detector at the EIC Stony Brook University (CFNS)4th AI4EIC Workshop 2 Physics Motivations at the EIC e− h •Semi-Inclusive Deep Inelastic Scattering •Production of hadrons in final-state" •Provide information on: " the fragmentation process (hadronization) the hadronic structure → → p/A Particle Identification detectors are crucial Stony Brook University (CFNS)4th AI4EIC Workshop 3 p K π The pfRICH Concept e− p/A θc∼θ2 sat − 1 n m2 p2 θc Particle Identification •Charged particle → emits Cherenkov photons at angle •Photons project onto photodetectors → form a ring Ring radius •Measuring ring size → deduce → particle mass θc → ∝tan θc θc Detection Principle The pfRICH will provide > 3σ π/K separations for momentum up to 7 GeV/c for –3.5 < η < -1.5 Stony Brook University (CFNS)4th AI4EIC Workshop 4 p K π The pfRICH Concept e− p/A θc∼θ2 sat − 1 n m2 p2 θc Particle Identification Can we use machine learning to improve particle identification? Stony Brook University (CFNS)4th AI4EIC Workshop 5 The approach π p K θc Physics Can we use machine learning to improve particle identification?: Yes! pfRICH PID π p K x1 x2 x3 xi An ideal use case for AI/ML, since the signal is well defined and fully understood AI/ML model Stony Brook University (CFNS)4th AI4EIC Workshop 6 x1 x2 x3 xi K p Training f(∑ i wixi+b) Model training π Standalone ePIC pfRICH GEANT4: Timing, hits position, momentum … More (good) data → better training AI/ML model XGBoost Gradient-boosted hybrid Diffusion model (ongoing) Stony Brook University (CFNS)4th AI4EIC Workshop 7 π x1 x2 x3 xi K p Training Data π p K Particle Identification f(∑ i wixi+b) Model inference AI/ML model Stony Brook University (CFNS)4th AI4EIC Workshop 8 Results Feature importanceConfusion matrix