Self-Organizing Maps
Segmented Overdetermined Nonlinear Independent Component Analysis for Online Neural Filtering
SBRN '08 Proceedings of the 2008 10th Brazilian Symposium on Neural Networks
Source separation in post-nonlinear mixtures
IEEE Transactions on Signal Processing
Hi-index | 0.00 |
High-energy detectors operating in particle collider experiments typically require efficient online filtering to guarantee that most of the background noise will be rejected and valuable information will not be lost. Among these types of detectors, calorimeters play an important role as they measure the energy of the incoming particles. In practical designs, calorimeter exhibit some sort of nonlinear behavior. In this paper, nonlinear independent component analysis (NLICA) methods are applied to extract relevant features from calorimeter data and produce high-efficient neural particle discriminators for online filtering operation. The study is performed for ATLAS experiment, one of the main detectors of the Large Hadron Collider (LHC), which is a last generation particle collider currently under operational tests. A performance comparison between different NLICA algorithms (PNL, SOM and Local ICA) is presented and it is shown that all outperform the baseline discriminator, that is based on classical statistical approach.