Rekonstrukcija mlazova pomoću metoda strojnog učenja

master's thesis

• Title (english): Jet reconstruction with machine learning methods
• Author: Martin Bajzek
• Mentor: Nikola Poljak (mentor)
• Committee member: Nikola Poljak (predsjednik povjerenstva)
• Committee member: Mirko Planinić (član povjerenstva)
• Committee member: Davor Horvatić (član povjerenstva)
• Committee member: Krešimir Kumerički (član povjerenstva)
• Committee member: Mihael Makek (član povjerenstva)
• Granter: University of Zagreb; Faculty of Science; (Department of Physics); Zagreb
• Defense date and country: 2021-09-28, Croatia
• Scientific / art field, discipline and subdiscipline: NATURAL SCIENCES; Physics

Abstract

Rekonstrukcija cijelog mlaza u modernoj fizici problematična je zbog detektorskih efekata i nepotpunog poznavanja hadronizacije. Rekonstrukcija se temelji na simulacijama događaja i simuliranom odazivu detektora iz kojeg se onda određuje efikasnost rekonstrukcije pojedinog kanala koji se razmatra. Efikasnost se tada uračunava u mjerene podatke stvarnih događaja, za koje se pretpostavlja da prate iste distribucije kao i simulirani događaji. U ovom radu bi se napravio novi, pojednostavljeni simulator mlazova koji bi usmjeravali na N x N polje koje predstavlja detektor. U svakom elementu polja, za pojednostavljeni snop, zapisao bi se četverovektor impulsa i energije te bi cjelokupno polje predstavljalo fizikalni snop. Realna slika snopa tad bi se dobila uvođenjem detektorskih efekata na polje. Konačno, vidjet ćemo je li pomoću metoda strojnog učenja moguće rekonstruirati fizikalni snop iz njegove realne slike, te je li efikasnost takve rekonstrukcije usporediva s efikasnostima modernih metoda.

Abstract (english)

Jet reconstruction in modern physics poses problems due to detector effects and incomplete knowledge of hadronisation. The reconstruction is based on simulated events and simulated responses of the detector from which the efficiency of the reconstruction in individual channels is determined. Efficiency is then included in the measurement data of real events, which is assumed to follow the same distributions as the simulated events. In this paper, we generate a new simplified jet simulator that would be directed onto the N x N field represented by the detector. In each element of the field, for a simplified beam, an energy-momentum four-vector would be written and the whole field would represent a physical beam. Realistic image of the beam would then be obtained by introducing detector effects on the field. Finally, we’ll see if it is possible to reconstruct the physical jet from its images by using machine learning method and whether the efficiency of such reconstruction is comparable to the efficiencies of modern methods.

• Keywords: strojno učenje; detektori; Keras
• Keywords (english): machine learning; detectors; Keras
• Language: croatian
• URN:NBN: urn:nbn:hr:217:510280
• Study programme: Title: Physics; specializations in: Research; Course: Research; Study programme type: university; Study level: integrated undergraduate and graduate; Academic / professional title: magistar/magistra fizike (magistar/magistra fizike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2021-11-10 13:07:17