Samoadaptivno mrežno usmjeravanje

master's thesis

• Title (english): Selfadaptive network routing
• Author: Robert Čupković
• Mentor: Dragan Jevtić (mentor)
• Committee member: Dragan Jevtić (predsjednik povjerenstva)
• Committee member: Marin Vuković (član povjerenstva)
• Committee member: Miljenko Mikuc (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-05, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Problem usmjeravanja informacija zahtjeva inteligentno rješenje uslijed konstantnih promjena u telekomunikacijskoj mreži. Jedna od mogućnosti je primjena algoritma iz područja poticanog učenja, podvrste strojnog učenja u kojoj softverski agent na temelju pokušaja i pogrešaka upoznaje svoju okolinu koja ga poticajima usmjerava prema cilju. Često korišteni algoritam poticanog učenja je Q-učenje. Za pohranu Q vrijednosti koristi se Q tablica, no nju se može zamijeniti umjetnom neuronskom mrežom ako je prihvatljivo izgubiti dio preciznosti. U području mrežnog usmjeravanja Q-algoritmom, neka od rješenja su Q-usmjeravanje, Full-Echo Q-usmjeravanje i Dual Reinforcement Q-usmjeravanje. Pri implementaciji laboratorijskog modela spojeno je Full-Echo Q-usmjeravanje s dubokim Q-učenjem te je napravljena validacija istog. Dobiveni su zadovoljavajući rezultati i pokazalo se da odabrani pristup ima prostora za napredak u budućnosti.

Abstract (english)

The problem of routing information requires an intelligent solution due to constant changes in a telecommunications network. One possibility is applying an algorithm from the field of reinforcement learning, an area of machine learning in which a software agent learns about his environment using the method of trial and error, all the while the environment guides him towards the goal. Q-learning is an example of an often used reinforcement learning algorithm. A table is used in order to store Q values, however it can be replaced with an artificial neural network if it's acceptable to lose some accuracy. In the field of network routing, some Q-algorithm-based solutions include Q-routing, Full-Echo Q-routing and Dual Reinforcement Q-routing. The implementation of the laboratory model combines Full-Echo Q-routing with Deep Q-learning and includes the validation of said model. The ending results are satisfactory and it's shown there's room for improvement in the future using the described approach.

• Keywords: mrežno usmjeravanje; adaptacija; poticano učenje; Q-učenje; umjetna neuronska mreža; Full-Echo; evaluacija algoritma
• Keywords (english): network routing; adaptation; reinforcement learning; Q-learning; artificial neural network; Full-Echo; algorithm evaluation
• Language: croatian
• URN:NBN: urn:nbn:hr:168:764097
• 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: 2020-01-07 17:27:11