| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

Show document Help

Title:Analiza ranljivosti in varnostni vidiki zastrupljanja podatkov pri učenju velikih jezikovnih modelov : magistrsko delo
Authors:ID Dimkovska, Vanja (Author)
ID Karakatič, Sašo (Mentor) More about this mentor... New window
Files:.pdf MAG_Dimkovska_Vanja_2025.pdf (3,17 MB)
MD5: 93DFFAB51AAB9F98F5944791E8E9FBBA
 
Language:Slovenian
Work type:Master's thesis/paper
Typology:2.09 - Master's Thesis
Organization:FERI - Faculty of Electrical Engineering and Computer Science
Abstract:Predmet magistrske naloge je analiza ranljivosti velikih jezikovnih modelov ob zastrupljanju podatkov v fazi učenja. Delo se osredotoča na preučevanje, kako različni deleži in vrste zastrupljenih podatkov vplivajo na kakovost odgovorov modelov, pri čemer je poudarek na ciljnem in tematsko nepovezanem zastrupljanju. Raziskava temelji na eksperimentalnem pristopu z uporabo medicinskih učnih nizov ter vključuje analizo sprememb v vedenju modelov pri različnih scenarijih zastrupljanja.
Keywords:veliki jezikovni modeli, zastrupljanje podatkov, varnost umetne inteligence, kakovost odgovorov, medicinski podatki
Place of publishing:Maribor
Place of performance:Maribor
Publisher:[V. Dimkovska]
Year of publishing:2025
Number of pages:1 spletni vir (1 datoteka PDF (VIII, 69 str.))
PID:20.500.12556/DKUM-93623 New window
UDC:004.6.056:004.85(043.2)
COBISS.SI-ID:245839875 New window
Publication date in DKUM:13.08.2025
Views:208
Downloads:48
Metadata:XML DC-XML DC-RDF
Categories:KTFMB - FERI
:
Copy citation
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:Bookmark and Share



Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Licences

License:CC BY-NC-ND 4.0, Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Link:http://creativecommons.org/licenses/by-nc-nd/4.0/
Description:The most restrictive Creative Commons license. This only allows people to download and share the work for no commercial gain and for no other purposes.
Licensing start date:08.07.2025

Secondary language

Language:English
Title:Vulnerability analysis and security aspects of data poisoning in large language model training
Abstract:This master's thesis focuses on analyzing vulnerabilities in large language models resulting from data poisoning during the training phase. The work focuses on examining how different percentages and types of poisoned data affect the quality of model responses, with an emphasis on targeted and off-topic poisoning. The research is based on an experimental approach using medical training datasets and includes an analysis of behavioral changes in models under various poisoning scenarios.
Keywords:large language models, data poisoning, AI security, answer quality, medical datasets


Comments

Leave comment

You must log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica