S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop
Machine Lea ning o he p RICH
Pa icle Iden i ica ion subsys em
D.H Dongwi, C-J. Naïm and L. Rhode
A i icial In elligence o he Elec on Ion Collide (2025)
Oc obe , 28, 2025
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 1
•A compac cen al de ec o wi h se e al subsys ems"
•He me ic co e age: –3.5 < η < 3.5 ( acking, calo ime y, pa icle iden i ica ion)
p/A
e
p RICH subsys em in backwa d egion
The ePIC de ec o a he EIC
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 2
Physics Mo i a ions a he EIC
e−
h
•Semi-Inclusi e Deep Inelas ic Sca e ing
•P oduc ion o had ons in inal-s a e"
•P o ide in o ma ion on: "
he agmen a ion p ocess (had oniza ion)
he had onic s uc u e
→
→
p/A
Pa icle Iden i ica ion de ec o s a e c ucial
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 3
p
K
π
The p RICH Concep
e−
p/A
θc∼θ2
sa −
1
n
m2
p2
θc
Pa icle Iden i ica ion
•Cha ged pa icle → emi s Che enko pho ons a angle
•Pho ons p ojec on o pho ode ec o s → o m a ing
Ring adius
•Measu ing ing size → deduce → pa icle mass
θc
→
∝ an θc
θc
De ec ion P inciple
The p RICH will p o ide > 3σ π/K sepa a ions o
momen um up o 7 GeV/c o –3.5 < η < -1.5
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 4
p
K
π
The p RICH Concep
e−
p/A
θc∼θ2
sa −
1
n
m2
p2
θc
Pa icle Iden i ica ion
Can we use machine lea ning o imp o e
pa icle iden i ica ion?
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 5
The app oach
π
p
K
θc
Physics
Can we use machine lea ning o imp o e pa icle iden i ica ion?: Yes!
p RICH PID
π
p
K
x1
x2
x3
xi
An ideal use case o AI/ML, since he signal is well de ined and ully unde s ood
AI/ML model
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 6
x1
x2
x3
xi
K
p
T aining
(∑
i
wixi+b)
Model aining
π
S andalone ePIC p RICH GEANT4:
Timing, hi s posi ion, momen um …
Mo e (good) da a → be e aining
AI/ML model
XGBoos
G adien -boos ed hyb id
Di usion model (ongoing)
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 7
π
x1
x2
x3
xi
K
p
T aining
Da a
π
p
K
Pa icle Iden i ica ion
(∑
i
wixi+b)
Model in e ence
AI/ML model
S ony B ook Uni e si y (CFNS)4 h AI4EIC Wo kshop 8
Resul s
Fea u e impo anceCon usion ma ix
