Analiza utjecaja odabira hiperparametara na kolaborativno učenje

master's thesis

• Title (english): Analyzing the Influence of Hyperparameters on Federated Learning
• Author: Ivan Pikelj
• Mentor: Damir Pintar (mentor)
• Committee member: Damir Pintar (predsjednik povjerenstva)
• Committee member: Luka Humski (član povjerenstva)
• Committee member: Boris Vrdoljak (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-09, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Kolaborativno učenje je novi pristup treniranju modela umjetne inteligencije na distribuiranim podacima i uređajima. Ovakvim pristupom podaci ostaju pohranjeni lokalno i time se čuva privatnost. Treniranje modela u tako heterogenom sustavu dovodi do novih izazova različitih od već dobro utvrđenih metoda. U radu su opisane uobičajene metode kolaborativnog procesa, a dokaz koncepta ukazuje na utjecaj hiperparametara na proces. Dokaz koncepta pokazuje izazove koji se mogu pojaviti u procesu treniranja nad akademskim podacima kategorizacije MNIST brojeva i kako ih se može ublažiti dobrim odabirom hiperparametara, ovisno o problemu.

Abstract (english)

Federated learning is a new approach to train models of artificial intelligence on distributed data and devices. With this approach data can be left stored locally and will preserve privacy. Training models in such elaborate system induces new challenges in contrast to well established methods. Paper describes common methods to achieve such process as well gives proof of concept to demonstrate significance of hyperparameters on the process. Proof of concept demonstrates challenges models might occur in process of training on academic data of categorizing MNIST numbers and how it can be alleviated via good hyperparameter selection depending on problem.

• Keywords: Kolaborativno učenje; Federirano učenje; strojno učenje; distribuirani sustav
• Keywords (english): Federated Learning; Collaborative learning; machine learning; distributed system
• Language: croatian
• URN:NBN: urn:nbn:hr:168:102235
• 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 10:36:31