Detekcija anomalija u LHC-u zasnovana na autoenkoderima

master's thesis

• Title (english): Anomaly detection at LHC based on autoencoders
• Author: Ema Puljak
• Mentor: Zvonko Kostanjčar (mentor)
• Committee member: Zvonko Kostanjčar (predsjednik povjerenstva)
• Committee member: Tomislav Petković (član povjerenstva)
• Committee member: Stjepan Begušić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-09-17, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Potraga za novim pojavama u fizici koje su izvan Standardnog modela fizike jedan je od glavnih razloga za projektiranje i izradu Velikog Hadronskog Sudarača (engl. Large Hadron Collider) u CERN -u. Ovaj rad opisuje razvoj algoritama nenadziranog učenja za otkrivanje anomalija, osobito standardnih i varijacijskih autokodera, te strategije za njihovu primjenu u ograničenom okruženju sustava za filtriranje događaja koji djeluje u stvarnom vremenu u LHC eksperimentu. Takav model bio bi neovisan o prirodi anomalije, pa bi bio nepristran pri odabiru novih pojava u fizici. Dvije vrste autoenkodera s potpuno povezanom arhitekturom neuronske mreže uspoređuju se u smislu učinkovitosti otkrivanja anomalija i performansi u pogledu kašnjenja i korištenja resursa nakon što su raspoređene na primjeru programirljivih logičkih sklopova (engl. Field- programmable gate array) tipičnog sustava za odabir događaja na LHC-u. Također, pokazano je da model nenadziranog može učinkovito postići dobre performanse otkrivanja anomalija unutar strogo zadanih dopuštenih intervala kašnjenja i ograničenja resursa koje nameće FPGA okruženje.

Abstract (english)

The search for new physics beyond the Standard Model is one of the main reasons for the creation of the Large Hadron Collider (LHC) at CERN. This thesis describes the development of unsupervised learning algorithms for anomaly detection, particu- larly standard and variational autoencoders, and the strategies for their deployment in the restricted environment of the online event filtering system at an LHC experiment. Such a model would be independent of the nature of the anomaly, thus being unbiased while selecting new physics signatures. Two types of autoencoders with fully con- nected neural network architecture are compared in terms of anomaly detection effi- ciency, and performance regarding latency and resource usage after being deployed on an example field-programmable gate array (FPGA) of a typical event selection system at the LHC. It is also demonstrated that the unsupervised learning model can effec- tively achieve good anomaly detection performance within strict latency and resource budgets imposed by the FPGA environment.

• Keywords: Otkrivanje anomalija; autokoder; varijacijski autokoder; Veliki Hadronski Sudarač; Kompaktni Muonski Solenoid; programirljivi logički sklopovi; kvantizacija; interval kašnjenja (engl. latency); korištenje resursa
• Keywords (english): Anomaly detection; Autoencoder; Variational Autoencoder; Large Hadron Collider; Compact Muon Solenoid; Field Programmable Gate Arrays; Level-1 Trigger; Quantization; Pruning; Latency; Resource utilization
• Language: english
• URN:NBN: urn:nbn:hr:168:104372
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 13:36:37