Abstract

Abstract

Machine Learning in Science and Engineering / Advanced machine-learning solutions in LHCb operations and data analysis.

The LHCb experiment exploits the vast quantities of beauty and charm hadrons produced in the LHC’s 40 MHz proton-proton collisions. LHCb’s unique physics programme is reflected in unique challenges in trigger, event selection and data analysis that are fundamentally different to those of other LHC experiments. To maximise the signal content in the data written to storage (currently 12k events per second), an increasing amount of event selection and processing takes place online, at trigger level, before any events are saved. LHCb is the only LHC experiment to have full, offline-quality level data available at this stage. The need to process large amounts of data within the constraints of the online and offline computing resources requires advanced data-analysis techniques. LHCb takes data at rates significantly higher than design, thanks also to purpose-developed machine-learning (ML) solutions. Such solutions are applied to an increasing class of essential online and offline tasks, including more precise and faster real-time classification of interesting events, smarter detector-performance calibrations, and more precise, efficient, and unbiased offline characterization of reconstructed events. This talk provides an overview recent original ML applications in the trigger, operations, and analysis of LHCb data in 2015-2016 and discusses ongoing and future developments.