Speaker
Description
Physicists continue to invest significant effort in the search for dark matter using increasingly large and sensitive detectors. ARGO is a next generation liquid argon (LAr) experiment designed to achieve enhanced sensitivity through advanced photodetection and large-scale instrumentation. The detector design under study employs Single Photon Avalanche Diodes (SPADs) with digital readout over a total instrumented surface of approximately 200 m², requiring the simultaneous handling of millions of data channels. This scale presents significant challenges for data acquisition systems in terms of power consumption, cabling complexity, and data storage, motivating the use of real-time data processing and reduction near the detector.
In this work, we investigate the use of real-time machine-learning (ML) techniques as part of the ARGO data acquisition chain. One convolutional neural network model (CNN) classifies particle interactions, while another reconstructs event position. Performance is evaluated using particle identification accuracy and position reconstruction error distributions. Ongoing work explores integrating both tasks into a unified CNN model to improve performance and reduce edge computing requirements.
| Keyword-1 | Edge Machine Learning |
|---|---|
| Keyword-2 | Dark Matter Instrumentation |
| Keyword-3 | Real Time Signal Processing |