Task leads: Noemi Calace (CERN), Stephanie Majewski (Oregon)

The goal of this task is the development of an algorithmic solution for the ATLAS Event Filter track reconstruction, employing optimal classical numerical and Machine Learning techniques, and to deploy it on the most suitable hardware architecture. Machine Learning approaches to tracking, as Graph Neural Networks, will be investigated to replace parts of the (or possibly the full) classical numerical algorithm chain. The aim is to optimize the physics and processing performance of the track reconstruction and to investigate the potential of porting parts of the tracking chain on systems with co-processors like GPUs and FPGAs. This task will implement the tools developed in T1.2 (Development framework towards fast inference of complex network architectures on LHC online systems) and in T1.7 (Framework integration of accelerators) and provide feedback on their performance for further optimization.