Artificial Neural Networks for reconstruction of energy losses in dead materials between barrel LAr and Tile calorimeters: exploration and results

In the course of computational experiments with Monte-Carlo events for ATLAS Combined Test Beam 2004 setup Artificial Neural Networks (ANN) technique was applied for reconstruction of energy losses in dead materials between barrel LAr and Tile calorimeters (Edm). The constructed ANN procedures exploit as their input vectors the information content of different sets of variables (parameters) which describe particular features of the hadronic shower of an event in ATLAS calorimeters. It was shown that application of ANN procedures allows one to reach 40% reduction of the Edm reconstruction error compared to the conventional procedure used in ATLAS collaboration. Impact of various features of a shower on the precision of $Edm$ reconstruction is presented in detail. It was found that longitudinal shower profile information brings greater improvement in $Edm$ reconstruction accuracy than cell energies information in LAr3 and Tile1 samplings.