Generiranje umjetnih podataka za testiranje i verifikaciju modela strojnog učenja

master's thesis

• Title (english): Generating Data for Testing and Verification of Machine Learning Models
• Author: Nikola Buhiniček
• Mentor: Krešimir Pripužić (mentor)
• Committee member: Krešimir Pripužić (predsjednik povjerenstva)
• Committee member: Siniša Sovilj (član povjerenstva)
• Committee member: Jurica Babić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-09, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Testiranje i verifikacija modela strojnog učenja uvelike ovisi o podacima nad kojima se ti procesi izvode. Za određene probleme strojnog učenja teško je doći do skupova podataka koji zadovoljavaju sve njihove potrebe. Ograničenja tih podataka mogu biti razna, poput nedostatne količine skupa podataka, ograničenja zaštite privatnosti korisnika, statistička ograničenja skupa podataka i mnoga druga. Korištenjem umjetno generiranih podataka se smanjuju, ili čak u potpunosti uklanjaju, prethodno spomenuta ograničenja. Tako se smanjuju pogreške modela koje su nastale kao posljedica korištenja neprikladnih skupova podataka. U ovom radu istražit će se najčešće korišteni modeli u području generiranja umjetnih podataka, te će biti predložena i implementirana četiri generatora umjetnih podataka za odabrani problem strojnog učenja koristeći programski jezik Python. Generirani skupovi podataka će biti prokomentirani s obzirom na problem koji se njima rješava.

Abstract (english)

Testing and verification of machine learning models largely depends on the data over which these processes are performed. For certain machine learning problems, it is difficult to come up with datasets that meet all their needs. The limitations of real datasets can be various, such as insufficient amount of data, limitations of the dataset to protect user privacy, statistical limitations of the dataset and many others. The use of synthetically generated data reduces, or even eliminates these limitations which reduces model errors that occur because of using inappropriate datasets. In this paper, existing models used in the field of synthetic data generation were researched and four synthetic data generators, for a selected machine learning problem, were proposed and implemented using the Python programming language. The results obtained were commented with respect to the problem which was being solved.

• Keywords: generiranje umjetnih podataka; strojno učenje; generativni modeli; generativne kontradiktorne mreže; Python; WGAN; TGAN
• Keywords (english): synthetic data generation; machine learning; generative models; generative adversarial networks; Python; WGAN; TGAN
• Language: croatian
• URN:NBN: urn:nbn:hr:168:841456
• 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 14:38:40