Work

Regularisoitu koneoppimismalli hiukkassuihkujen luokitteluun

Year: 2024
Type: dissertation
Abstract: Particle physics research helps to understand some of the most fundamental parts of the universe. Many of the elementary particles do not exist in normal conditions but can be artificially created and studied using particle accelerators like the Large Hadron Collider (LHC) of the European Organization for Nuclear Physics (CERN). The future upgrade will increase the luminosity of the LHC, meaning vast amounts of data will be produced. In order to select only a small fraction of interesting data for further analysis, trigger algorithms are utilized. The triggers utilize custom electronics like Field-Programmable Gate Arrays (FPGAs) that enable fast decision-making but have restricted hardware capabilities. This study utilizes regularization, quantization, pruning, and hyperparameter optimization to implement a machine learning model for trigger-level jet classification on an FPGA platform. The results indicate that with the multi-objective hyperparameter optimization method, it is possible to retain the performance achieved with a baseline 4-layer multilayer perceptron model, with significantly reduced model resource usage.
Author: Ylijoki, Joona
Language: English
Open Access status: green
OpenAlex ID: W7014891483
Cites: 43
Cited by:
Related to: 10