PoS - Proceedings of Science
Volume 314 - The European Physical Society Conference on High Energy Physics (EPS-HEP2017) - Detector R&D and Data Handling (Parallel Session). Conveners: Paula Collins; Katja Kruger. Scientific Secretary: Enrico Conti.
Machine and deep learning techniques in heavy-ion collisions with ALICE
R. Haake*  on behalf of the ALICE collaboration
Full text: pdf
Pre-published on: October 19, 2017
Published on: March 20, 2018
Abstract
Over the last years, machine learning tools have been successfully applied to a wealth of problems in high-energy physics.
A typical example is the classification of physics objects.
Supervised machine learning methods allow for significant improvements in classification problems by taking into account observable correlations and by learning the optimal selection from examples, e.g. from Monte Carlo simulations.
Even more promising is the usage of deep learning techniques. Methods like deep convolutional networks might be able to catch features from low-level parameters that are not exploited by default cut-based methods.

These ideas could be particularly beneficial for measurements in heavy-ion collisions, because of the very large multiplicities. Indeed, machine learning methods potentially
perform much better in systems with a large number of degrees of freedom compared to cut-based methods. Moreover, many key heavy-ion observables are most interesting at low transverse momentum where the underlying event is dominant and the signal-to-noise ratio is quite low.

In this work, recent developments of machine- and deep learning applications in heavy-ion collisions with ALICE will be presented, with focus on a deep learning-based b-jet tagging approach and the measurement of low-mass dielectrons. While the b-jet tagger is based on a mixture of shallow fully-connected and deep convolutional networks, the low-mass dielectron measurement uses gradient boosting and shallow neural networks. Both methods are very promising compared to default cut-based methods.
DOI: https://doi.org/10.22323/1.314.0498
How to cite

Metadata are provided both in "article" format (very similar to INSPIRE) as this helps creating very compact bibliographies which can be beneficial to authors and readers, and in "proceeding" format which is more detailed and complete.

Open Access
Creative Commons LicenseCopyright owned by the author(s) under the term of the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.