EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A comprehensive analysis of objective functions in adversarial training

Ammar Al-Najjar

Pollack Periodica · 09.11.2026 · 21 (3) · 17-22

DOI: 10.1556/606.2025.01231 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Pollack Periodica, erschienen am 09.11.2026. Verfasst von Ammar Al-Najjar. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: European Union project.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This study explores adversarial robustness in machine learning by analyzing objective functions in the adversarial training defense. This study addresses its min-max optimization problem, leveraging knowledge distillation, self-alignment, and ground truth “hard-labels” losses to assess their effectiveness. By examining combinations of seven key terms with benign and adversarial inputs, alongside hard-labels, analysis demonstrates how various objective functions enhance model resilience. The results reveal top-performing combinations, providing insights for optimizing adversarial training to achieve improved robustness against adversarial attacks.

Förderung

  • European Union project

Zitieren

APA

Al-Najjar, A. (2026). A comprehensive analysis of objective functions in adversarial training. Pollack Periodica, 21(3), 17-22. https://doi.org/10.1556/606.2025.01231

Vancouver

Al-Najjar A. A comprehensive analysis of objective functions in adversarial training. Pollack Periodica. 2026;21(3):17-22. doi:10.1556/606.2025.01231

BibTeX
@article{AlNajjar2026,
  author = {Al-Najjar, Ammar},
  title = {A comprehensive analysis of objective functions in adversarial training},
  journal = {Pollack Periodica},
  year = {2026},
  volume = {21},
  number = {3},
  pages = {17--22},
  doi = {10.1556/606.2025.01231}
}
Kostenlos starten →