Machine learning in CMS

Abstract

Advanced machine learning (ML) methods are increasingly used in CMS physics analyses to maximize the sensitivity of a wide range of measurements. The landscape is diverse in terms of both methods and applications. Deep learning methods, from recurrent long short-term memory (LSTM) architectures for classification tasks to deep autoencoders for data quality monitoring, have greatly improved the physics results delivered from the CMS experiment. Algorithms are developed both for collaboration-wide use as well as for individual physics analyses. Results from CMS, such as the measurement of the Higgs boson’s properties in the diphoton decay channel, exploit a variety of ML algorithms to reduce uncertainties on measurements.