MACHINE LEARNING AND SOFTWARE SOLUTIONS FOR DATA QUALITY ASSESSMENT IN CERN’S ATLAS EXPERIMENT

The Large Hadron Collider (LHC) is home to multiple particle physics experiments designed to verify the standard model and push our understanding of the universe to its limits. The ATLAS detector is one of the large general-purpose experiments that make use of the LHC and generates a significant amount of data as part of its regular operations. Prior to physics analysis, this data is cleaned through a data assessment process which involves significant operator resources. With the evolution of the field of machine learning and anomaly detection, there is great opportunity to upgrade the ATLAS Data Quality Monitoring Framework to include automated, machine learning based solutions to reduce operator requirements and improve data quality for physics analysis. This thesis provides an infrastructure, theoretical foundation and a unique machine learning approach to automate this process. It accomplishes this by combining 2 heavily documented algorithms (Autoencoders and DBScan) and organizing the dataset around geometric descriptor features. The results of this work are released as code and software solutions for the benefit of current and future data quality assessment, research, and collaborations in the ATLAS experiment