Kategorizacija podataka iz mjerenja 12C(n,p) i 12C(n,d) reakcija primjenom metoda strojnog učenja

master's thesis

• Title (english): Categorization of data from measurements of 12C(n,p) and 12C(n,d) reactions using machine learning methods
• Author: Tadej Barović
• Mentor: Petar Žugec (mentor)
• Committee member: Petar Žugec (predsjednik povjerenstva)
• Committee member: Ozren Perše (član povjerenstva)
• Committee member: Ivica Friščić (član povjerenstva)
• Committee member: Boris Muha (član povjerenstva)
• Committee member: Darko Androić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Mathematics); Zagreb
• Defense date and country: 2022-07-21, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Mathematics
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Physics

Abstract

Cilj ovog diplomskog rada jest usporediti utjecaj različitih transformacija podataka na efektivnost tj. točnost klasifikatora reakcija \(^{12}\mathrm{C}(\mathrm{n,p})^{12}\)B i \(^{12}\mathrm{C(n,d)}^{11}\mathrm{B}\). Kao klasifikator koristit ćemo neuronsku mrežu. Kontekst ovog rada je eksperiment proveden u n TOF postrojenju na CERN-u. Tijekom obrade eksperimentalnih podataka rekonstruirat će se udarni presjek spomenutih reakcije preko podataka mjerenih silicijskim teleskopom. Mjerenje koje se temelji na \( \Delta\)E − E tehnici simulirali smo u GEANT4 programskom paketu. Simulirane podatke obradili smo i transformirali Box-Cox transformacijom i vokselizacijom. Neuronske mreže implementirali smo na izvornim i transformiranim podatcima. Box-Cox transformacija optimalna je u ovom kontekstu. Svim mrežama točnost naglo pada na energijama neutrona iznad 25 MeV. Takav zaključak u skladu je sa preliminarnim očekivanjima iz prijedloga mjerenja \(^{12}\mathrm{C}(\mathrm{n,p})^{12}\)B i \(^{12}\mathrm{C(n,d)}^{11}\mathrm{B}\) reakcija na n_TOF postrojenju

Abstract (english)

In this thesis we aim to compare the effectiveness of various transformation data on the accuracy of the \(^{12}\mathrm{C}(\mathrm{n,p})^{12}\)B and \(^{12}\mathrm{C(n,d)}^{11}\mathrm{B}\) reaction classifier – the neural network. The base of our paper is an experiment carried out in the n_TOF facility at CERN. In the experiment, the cross sections of the mentioned reactions are reconstructed using the data measured by a silicon telescope. We simulated the measurement, which is based on the \( \Delta\)E − E technique, in the GEANT4 software. We analyzed and transformed the simulated data using Box-Cox transformation and voxelization. The study also implemented neural networks on non-transformed data, along with implementation on each of the transformations. We conclude that the Box-Cox transformation is optimal in this context. For all networks, the accuracy drops sharply above 25 MeV of input neutron energy, confirming the previously set assumptions in the proposal of \(^{12}\mathrm{C}(\mathrm{n,p})^{12}\)B and \(^{12}\mathrm{C(n,d)}^{11}\mathrm{B}\) reaction measurement.

• Keywords: klasifikator reakcija; neuronska mreža; silicijski teleskop; GEANT4 programski paket; Box-Cox transformacija; Box-Cox vokselizacija
• Keywords (english): reaction classifier; neural network; silicon telescope; GEANT4 software; Box-Cox transformation; Box-Cox voxelization
• Language: croatian
• URN:NBN: urn:nbn:hr:217:366014
• Study programme: Title: Mathematics and Physics Education; specializations in: Mathematics and Physics Education; Course: Mathematics and Physics Education; Study programme type: university; Study level: integrated undergraduate and graduate; Academic / professional title: magistar/magistra edukacije matematike i fizike (magistar/magistra edukacije matematike i fizike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2022-09-12 11:34:49