Adversarial Attacks and Robustness in AI: Methods, Empirical Analysis, and Defense Mechanisms

Chenglin Song

2025

Abstract

Adversarial attacks pose significant threats to modern artificial intelligence (AI) systems by introducing subtle perturbations into input data that can drastically alter model predictions. These attacks have serious implications in safety-critical applications such as autonomous driving and healthcare, where reliability and robustness are essential. In addition to computer vision systems, adversarial vulnerabilities have been observed in natural language processing and speech recognition, further highlighting the broad scope of this issue. This paper provides an integrative review of adversarial attack generation techniques, discusses empirical findings on AI robustness, and surveys existing defense mechanisms. Through an examination of state-of-the-art research, current limitations are highlighted, and directions for developing more resilient AI models are proposed. Practical considerations and potential future applications are also outlined with the goal of informing both theoretical inquiry and real-world deployment strategies. Recent studies have further expanded on these topics by emphasizing enhanced adversarial training methods and layered defense architectures, which are also discussed in the context of new empirical evidence.

Paper Citation

in Harvard Style

Song C. (2025). Adversarial Attacks and Robustness in AI: Methods, Empirical Analysis, and Defense Mechanisms. In Proceedings of the 2nd International Conference on Innovations in Applied Mathematics, Physics, and Astronomy - Volume 1: IAMPA; ISBN 978-989-758-774-0, SciTePress, pages 402-406. DOI: 10.5220/0013826700004708

in Bibtex Style

@conference{iampa25,
author={Chenglin Song},
title={Adversarial Attacks and Robustness in AI: Methods, Empirical Analysis, and Defense Mechanisms},
booktitle={Proceedings of the 2nd International Conference on Innovations in Applied Mathematics, Physics, and Astronomy - Volume 1: IAMPA},
year={2025},
pages={402-406},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013826700004708},
isbn={978-989-758-774-0},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 2nd International Conference on Innovations in Applied Mathematics, Physics, and Astronomy - Volume 1: IAMPA
TI - Adversarial Attacks and Robustness in AI: Methods, Empirical Analysis, and Defense Mechanisms
SN - 978-989-758-774-0
AU - Song C.
PY - 2025
SP - 402
EP - 406
DO - 10.5220/0013826700004708
PB - SciTePress