DC FieldValueLanguage
dc.contributor.advisorZseby, Tanja-
dc.contributor.authorBachl, Maximilian-
dc.date.accessioned2021-07-14T08:27:18Z-
dc.date.issued2021-
dc.date.submitted2021-06-
dc.identifier.citation<div class="csl-bib-body"> <div class="csl-entry">Bachl, M. (2021). <i>Machine learning methods for communication networks : characterization  and analysis of selected use cases</i> [Dissertation, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2021.84423</div> </div>-
dc.identifier.urihttps://doi.org/10.34726/hss.2021.84423-
dc.identifier.urihttp://hdl.handle.net/20.500.12708/18054-
dc.description.abstractWith machine learning and especially deep learning rising to prevalence in many domains such as computer vision and natural language processing, machine learning methods are also increasingly investigated for computer networks. This work shows that while machine learning has great potential in some areas of computer networking, the challenges are very different than those found in other domains for which learning methods such as deep learning are commonly adopted.Specifically, I show that particularities of machine learning for networking are:• the need that a machine learning solution interacts well with existing deployed technology such as other network protocols and legacy solutions.• the different nature and smaller quantity of features compared to, for example, computer vision.• the common occurrence of attacks since many networks are open to the outside world.• the requirement of fast processing.These characteristics are elaborated on by developing and analyzing machine learning approaches for• Congestion control: I use reinforcement learning to learn a competitive congestion control policy and develop a mechanism which can improve fairness between different flows in the Internet.• Active queue management: I use reinforcement learning and also classic approaches to develop queue managers which inspect traffic and smartly choose the optimal buffer size for a particular flow.• Intrusion detection: I develop methods to quantify adversarial threats and mitigate them and a method which lowers resource consumption by learning to focus more on network traffic which seems interesting.Overall, I show that machine learning solutions offer promising performance improvements over static human-developed solutions when the above particularities are considered.en
dc.formativ, 188 Blätter-
dc.languageEnglish-
dc.language.isoen-
dc.subjectMachine Learningen
dc.subjectDeep Learningen
dc.subjectNetworkingen
dc.subjectIntrusion Detectionen
dc.subjectCongestion Controlen
dc.titleMachine learning methods for communication networks : characterization and analysis of selected use casesen
dc.typeThesisen
dc.typeHochschulschriftde
dc.identifier.doi10.34726/hss.2021.84423-
dc.publisher.placeWien-
tuw.thesisinformationTechnische Universität Wien-
dc.contributor.assistantFabini, Joachim-
tuw.publication.orgunitE389 - Institute of Telecommunications-
dc.type.qualificationlevelDoctoral-
dc.identifier.libraryidAC16255285-
dc.description.numberOfPages188-
dc.thesistypeDissertationde
dc.thesistypeDissertationen
tuw.assistant.orcid0000-0002-8285-1591-
item.openairecristypehttp://purl.org/coar/resource_type/c_18cf-
item.openairecristypehttp://purl.org/coar/resource_type/c_18cf-
item.openaccessfulltextOpen Access-
item.openairetypeThesis-
item.openairetypeHochschulschrift-
item.fulltextwith Fulltext-
item.languageiso639-1en-
item.grantfulltextopen-
item.cerifentitytypePublications-
item.cerifentitytypePublications-
Appears in Collections:Thesis

Files in this item:


Page view(s)

36
checked on Sep 16, 2021

Download(s)

51
checked on Sep 16, 2021

Google ScholarTM

Check


Items in reposiTUm are protected by copyright, with all rights reserved, unless otherwise indicated.