Neural network techniques for charged particles reconstruction in ATLAS New Small Wheels.
The reconstruction of charged particles inside a detector is one of the most challenging computational problems for modern particle physics experiments. In particular, the extremely high rate of particles impinging in the detectors requires the development of innovative pattern recognition methods able to be efficient in highly dense environments. Due to the high luminosity values that LHC will reach during the Run-3 and the HL-LHC phase, the number of interactions per bunch crossing will increase, consequently increasing the detector's occupancy. For the traditional tracking algorithms used in high-energy physics, a too high occupancy constitutes a limit, both in terms of performances and processing time. New strategies for particle tracking based on machine learning techniques are presented, showing results on simulated samples and their application to the new ATLAS detectors, the New Small Wheels. In particular, the development of a DNN for position reconstruction and of a RNN for pattern finding will be discussed. These algorithms have great potential thanks to their capability to identify complex correlation in data and to parallelize operations on architectures such as FPGAs.