Deep Neural Networks applications in experimental physics data analyses

One of the head investigations at the LHC and ATLAS Experiment is related to the study of the Quark Gluon Plasma (QGP), formed in ultra-relativistic heavy-ion collisions. Due to its short lifetime and spa- tial limitation, the QGP properties are impossible to be measured directly, and therefore, indirect methods must be used. One method is related to the measured quantities of the collimated sprays of particles, so- called jets, generated immediately after the collision takes place. The particle shower develops in the QGP, so understanding the jet energy loss processes and the modification of the fragmentation functions is crucial to infer the properties of this state of matter. In particular, jets resulting from the bottom (b) quark fragmentation are expected to interact with the QGP differently from the other jets providing addi- tional information about the nature of the QGP so identifying them is a must. Distinguishing bottom-jets from charm- and light-jets produced in proton-proton collisions is difficult, but the huge environment of Pb+Pb collisions makes the task particularly challenging. In the ATLAS Experiment the Neural Net- works, such as the DL1 - Deep Learning 1, are the most promising tools to provide an efficient jet flavor discrimination and b-tagging.