Federalno učenje u ugradbenim sustavima

master's thesis

• Title (english): Federated Learning in Embedded Systems
• Author: Ivan Premužić
• Mentor: Igor Čavrak (mentor)
• Committee member: Igor Čavrak (predsjednik povjerenstva)
• Committee member: Hrvoje Mlinarić (član povjerenstva)
• Committee member: Ivana Bosnić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Federalnog učenje je kolaborativna, distribuirana metoda za učenje globalnog dijeljenog modela. Koristi računalne mogućnosti krajnjih uređaja (pametni telefoni, uređaji u ugradbenim sustavima, internetu stvari, računarstvu na rubu itd.) za obavljanje procesa strojnog učenja, pri čemu se za treniranje koriste lokalni podaci. Svaki krajnji uređaj (klijentski čvor) trenira lokalni model nad svojim podacima, a centralni čvor kreira globalni model na temelju lokalnih modela. Lokalni podaci svakog klijenta, koji su često privatni, nikada se ne dijele s nekom trećom stranom i na taj način je sačuvana privatnost klijenta. Kako bi se olakšalo istraživanje federalnog učenja razvijeni su programski okviri za federalno učenje, a jedan od njih je Flower. U ovom radu proučeno je područje federalnog učenja, te programski okvir Flower i njegove karakteristike i prednosti u odnosu na slične programske okvire. Implementiran je sustav za stvaranje zajedničkog modela temeljen na programskom okviru Flower, te su ispitane performanse sustava s obzirom na dostignuti kvalitetu zajedničkog globalnog modela i modela dobivenog metodom klasičnog centralnog strojnog učenja.

Abstract (english)

Federated learning (FL) is collaborative, distributed method of machine learning used to create a shared global model. It uses computing capabilities of end devices (smartphones, devices in embedded systems, IoT, edge computing etc.) for machine learning on their local data. Every end device (client node) trains a local model on its local data, while central node creates global shared model based on information from local models. Local data of each client, which is often private, is never shared with other party and privacy of clients is preserved. To make exploration and development of federated learning easier, several frameworks for federated learning have been developed, one of which is Flower. In this work the field of federated learning is described and Flower framework for federated learning is explored. Using Flower framework, system for creating a shared global model is developed and its performance is evaluated based on achieved quality of shared global model and model created using classic centralized method of machine learning.

• Keywords: Kolaborativno učenje; Distribuirano učenje; Federalno učenje; Duboko učenje; Flower; Internet stvari; Računarstvo na rubu
• Keywords (english): Collaborative learning; Distributed learning; Federated learning; Deep learning; Flower; Internet of things; Edge computing
• Language: croatian
• URN:NBN: urn:nbn:hr:168:929288
• 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: 2023-01-13 13:33:45