Diplomamunka

This repository contains the LaTeX source and final PDF for my 2021 master's thesis at Budapest University of Technology and Economics.

The thesis is written in Hungarian. The topic is still relevant: privacy risks in face recognition embeddings and possible mitigation methods.

Thesis

• Hungarian title: Gépi tanulási eljárásokkal generált adatok adatvédelmi vizsgálata
• Rough English description: Privacy analysis of data representations generated by machine learning methods
• Final PDF: Csarno_Tamas_Diplomamunka.pdf

Summary

The thesis focuses on facial recognition systems built on deep metric learning and the privacy risks of face embeddings.

Main questions covered in the work:

• what kinds of personal information can be inferred from face embeddings
• whether embeddings leak demographic attributes such as age, gender, and race
• how robust those signals are once encoded in the embedding space
• whether adversarial methods can reduce leakage
• which cryptographic approaches may help protect embeddings in practice

The core result is that face embeddings are not only useful for identification, but also carry additional sensitive information that can be extracted with strong accuracy using machine learning models.

Topics Covered

• deep metric learning
• siamese networks for face recognition
• generative models
• privacy analysis of embeddings
• attacker modeling
• adversarial mitigation ideas
• locality-sensitive hashing
• homomorphic encryption

Repository Layout

• Csarno_Tamas_Diplomamunka.pdf: final thesis PDF
• thesis.tex: main LaTeX entry point
• content/: thesis chapters
• figures/: figures used in the thesis
• bib/: bibliography
• include/: LaTeX preamble and thesis template pieces

Notes

• The thesis text is in Hungarian.
• This repository is meant primarily as an academic artifact and source archive.
• The LaTeX source is included in full, together with figures and bibliography.