The recent progress in computing and ad-hoc software has significantly simplified the access to machine learning techniques and numerical optimisation. In the LHC and its injector complex, a very diverse and inhomogeneous set of problems present the right observables types to be addressed with the classic or most cutting edge machine learning algorithms. In this talk, we introduce a set of techniques that have been applied to the CERN accelerator complex to solve problems that would have been otherwise impossible or very complicated to solve classically. Specifically, we will introduce the usage of computer vision techniques, time series analysis and physics aware neural networks. For all these algorithms and principles, we will present real applications to the LHC and its injectors.
Short bio Francesco Maria Velotti
Francesco obtained his MSc at Universita' del Sannio in Electronic engineering and his PhD at Ecole Polytechnique Fédérale de Lausanne in 2017 in accelerator physics with studies regarding the HL-LHC injection system and crystal-shadowing slow extraction. He is now a CERN staff member in the SY department and ABT group. Since 2018 he is one of the SPS supervisors and directly involved in the operation of the SPS. His research topics include slow extraction losses and spill quality optimisation, as well as machine learning applications to ABT and accelerator systems in general.
Massimo Giovannozzi / Participants: 63