HarmoniFL: Προσαρμοσμένο σε πόρους εργαλείο Federated Learning

Βλαχάκης, Νίκος; Vlachakis, Nikos

dc.contributor.author	Βλαχάκης, Νίκος	el
dc.contributor.author	Vlachakis, Nikos	en
dc.date.accessioned	2024-07-30T09:02:11Z
dc.date.available	2024-07-30T09:02:11Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/59976
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.27672
dc.rights	Default License
dc.subject	Federated Learning	en
dc.subject	Resource Adaptive	en
dc.subject	Distributed Computing	en
dc.subject	Privacy Preservation	en
dc.subject	Machine Learning	en
dc.subject	Ομοσπονδιακή Μάθηση	el
dc.subject	Διατήρηση Απορρήτου	el
dc.subject	Μηχανική Μάθηση	el
dc.subject	Κατανεμημένη Υπολογιστική	el
dc.subject	Προσαρμογή στους Πόρους	el
dc.title	HarmoniFL: Προσαρμοσμένο σε πόρους εργαλείο Federated Learning	el
dc.title	HarmoniFL: Resource-Adaptive Federated Learning Tool	en
heal.type	bachelorThesis
heal.classification	Artificial Intelligence and Machine Learning	en
heal.language	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2024-04-03
heal.abstract	In the realm of Federated Learning (FL), the presumption of uniform processing capacity among participating clients overlooks the reality of diverse client hardware. This diversity introduces system heterogeneity, leading to disparities in computational resources that ultimately compromise the efficient utilization of distributed data, thus hindering optimal learning outcomes. Addressing this challenge, this thesis introduces HarmoniFL, an open-source initiative engineered to navigate the complexities of resource heterogeneity within federated learning frameworks. By focusing on enhancing efficiency and inclusivity, HarmoniFL employs a dynamic client selection protocol that leverages real-time metrics such as CPU load, memory availability, and network bandwidth to optimize device participation in learning tasks. Adaptive strategies—ranging from data sampling adjustments to epoch reduction and batch size optimization for high-demand devices—address the challenges posed by resource diversity. Our experimental analysis aims at two primary goals: minimizing training duration for less capable devices and improving the accuracy of the aggregated global model. Results demonstrate that HarmoniFL effectively reduces training times and enhances model performance, underscoring its potential to foster more equitable device participation in federated learning tasks without sacrificing learning quality. The code repository can be found at https://github.com/NikosVlachakis/harmoni-fl.git.	en
heal.advisorName	Tsoumakos, Dimitrios	en
heal.committeeMemberName	Garofalakis, Minos	en
heal.committeeMemberName	Pneymatikatos, Dionysios	en
heal.committeeMemberName	Tsoumakos, Dimitrios	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών	el
heal.academicPublisherID	ntua
heal.numberOfPages	64 σ.	el
heal.fullTextAvailability	false