26TH INTERNATIONAL CONFERENCE ON COMPUTING IN HIGH ENERGY & NUCLEAR PHYSICS (CHEP2023)

Level-3 Electron trigger with Artificial Intelligence for CLAS12

Not scheduled

1h

Hampton Roads Ballroom and Foyer Area (Norfolk Waterside Marriott)

Poster Poster Poster Session

Speaker

Gavalian, Gagik (Jefferson Lab)

Description

With ever increasing data rates in modern physics experiments there is a need to
partially analyze incoming data real time. Increasing data volumes create a need
to filter data to write out only events that are useful for physics analysis.
In this work we present a Level-3 trigger developed using Artificial Intelligence
to select electron trigger events at data acquisition level. Convolutional Neural Network
used combined data from Drift Chambers and Electromagnetic Calorimeter to identify
electron trigger, as a result providing 99.9% accuracy and 94% purity
(compared to 96% accuracy and 40% purity of level-1 trigger). The online version
of the Level three trigger is running on Nvidia Tesla T4 GPUs, and is capable to keep up
with Data Acquisition rate (18 kHz).