Anomaly Detection Using Deep Sparse Autoencoders for CERN Particle Detector Data

Filip Široký

Bachelor's thesis

Abstract:

The certification of the Compact Muon Solenoid (CMS) particle detector data, as used for physics analysis, is a crucial task to ensure the quality of all physics results published by CERN. Currently, the certification conducted by human experts is labour intensive and can only be segmented on a long period of time basis. This contribution focuses on the design and prototype of an automated certification system assessing data quality on ~23 seconds of data taking basis. Anomalies caused by detector malfunctions or sub-optimal reconstruction are unpredictable and rarely occur, making it difficult to use standard supervised classification methods such as feedforward neural networks. We base our prototype on a semi-supervised model which employs deep sparse autoencoders. This approach has been qualified successfully on CMS data collected during the 2016 Large Hadron Collider (LHC) run: we demonstrate its ability to detect anomalies with high accuracy and low false positive rate compared to the manual certification by experts. A key advantage of this approach over other machine learning technologies is having great interpretability of the results, which can be further used to ascribe the origin of the problems in the data to a specific sub-detector or particle physics objects.

Abstract:

The certification of the Compact Muon Solenoid (CMS) particle detector data, as used for physics analysis, is a crucial task to ensure the quality of all physics results published by CERN. Currently, the certification conducted by human experts is labour intensive and can only be segmented on a long period of time basis. This contribution focuses on the design and prototype of an automated certification system assessing data quality on ~23 seconds of data taking basis. Anomalies caused by detector malfunctions or sub-optimal reconstruction are unpredictable and rarely occur, making it difficult to use standard supervised classification methods such as feedforward neural networks. We base our prototype on a semi-supervised model which employs deep sparse autoencoders. This approach has been qualified successfully on CMS data collected during the 2016 Large Hadron Collider (LHC) run: we demonstrate its ability to detect anomalies with high accuracy and low false positive rate compared to the manual certification by experts. A key advantage of this approach over other machine learning technologies is having great interpretability of the results, which can be further used to ascribe the origin of the problems in the data to a specific sub-detector or particle physics objects.

Keywords

machine learning, deep learning, neural networks, autoencoders, automation, particle detector, anomaly detection, big data, CERN, LHC, data science

Language used: English

Date on which the thesis was submitted / produced: 27. 5. 2019

Thesis defence

• Date of defence: 25. 6. 2019
• Supervisor: doc. RNDr. Petr Sojka, Ph.D.
• Reader: RNDr. Petr Novotný, Ph.D., Giovanni Franzoni, Ph.D.

Masaryk University

Faculty of Informatics

Bachelor programme / field:
Informatics / Mathematical Informatics