Neural network based cluster reconstruction in the ATLAS pixel detector

The ATLAS Pixel detector currently determining particle positions at 8 TeV proton-proton collisions is working with a dense track environment. Due to these tiny particle separations, shared cluster are produced. Thus, the aim of the NN implementation is to identify merged clusters and improve the particle position resolution. By combining many variables with non-linear correlations, the NN is ideal to estimate the number of particles passing through a cluster and each of their position and uncertainty. As a result of the NN reconstruction, the impact parameter improves by ~15% which indicates boosted prospects for physics analysis.