Speaker
Description
The ALICE TPC is the main tracking and PID detector used in the ALICE experiment at CERN. The online reconstruction is capable of handling dense tracking environments at data rates of 900 GB/s, with a GPU-based infrastructure, ideally suited for parallelizable machine learning applications.
The work to be presented concerns cluster finding, with the first-ever application of neural networks in ALICE online processing. Both a classification network for noise removal and a regression network for cluster property inference are presented. A 3D charge input to the cluster finding step marks a new approach taken for this challenge. The tuning of this algorithm for physics and computing performance is the major objectives for deployment purposes. Design optimizations of the network architecture, floating-point quantization, custom CUDA-streamed implementation using the ONNX Runtime framework and results from the first commissioning runs mark the cornerstones of this project. The achieved performance is a reduction in number of clusters of up to 18% with maintained or improved physics performance, demonstrated on simulated and real data. This includes tracking and matching efficiencies, track quality measurements, d$E$/d$x$ bandwidth investigations and invariant mass spectra of reconstructed $\Lambda, K^0_S$ and $D^0$ particles.
| I read the instructions above | Yes |
|---|