Obrane klasifikacijskih modela od zatrovanih skupova podataka

master's thesis

• Title (english): Defending classification models from poisoned datasets
• Author: Tin Reiter
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-11, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog rada je pokazati načine napada na duboke modele pomoću zatrovanih skupova podataka te istražiti kako ih obraniti od takvih napada. Želimo vidjeti koliko su efektivne trenutne metode obrane dubokih modela od takvih napada. U radu prvo opisujemo strukturu umjetnih neurona i dubokih modela. Zatim objašnjavamo različite načine napada pomoću zatrovanih skupova podataka na duboke modele. Nakon toga objašnjavamo razne načine obrana od tih napada. Na kraju prikazujemo eksperimentalne rezultate napada na duboke modele te efektivnost načina obrane protiv njih.

Abstract (english)

The goal of this paper is to demonstrate backdoor attacks on deep models and ways to mitigate them. We want to see how effective are the current defenses against backdoor attacks. In this paper we first describe the structure of artificial neurons and deep models. Then we describe the different types of defenses that can be used to mitigate backdoor attacks. In the end we show the experimental results gained from backdoor attacks on deep models and the effectiveness of the different methods to mitigate them.

• Keywords: duboko učenje; trovanje skupova podataka; obrana klasifikacijskih modela; duboki modeli
• Keywords (english): deep learning; backdoor attacks on deep models; mitigating backdoor attacks; deep models
• Language: croatian
• URN:NBN: urn:nbn:hr:168:034860
• 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: 2023-10-11 10:19:39