In High Energy Physics experiments Particle Flow (PFlow) algorithms are designed to provide an optimal reconstruction of the nature and kinematic properties of the particles produced within the detector acceptance during collisions. At the heart of PFlow algorithms [ref. 1] is the ability to distinguish the calorimeter energy deposits of neutral particles from those of charged particles, using the complementary measurements of charged particle tracking devices, to provide a superior measurement of the particle content and kinematics. We proposed a neural network based method which estimates the energy deposited by neutral and charged particles, in a given calorimeter cell, starting from the full energy distribution in the calorimeter and track information [ref. 2].
We compared the performance of four different neural networks. They are a pure convolutional network, the convolutional network with an UNet block, a graph network with a dynamic edge-convolution network, and deepset network. We show that both in terms of energy and direction estimation, the neural networks outperform the existing PFlow methods.
In the same article, we proposed a novel super-resolution technique that can estimate the high granularity shower profile, starting from the low-granularity detector information.
At the moment we are on our way to implement these techniques within ATLAS.
----------------------------------------------------------------------------------------