Sang-ho Lee, Youngho Cho, Journal of Internet Computing and Services, Vol. 25, No. 6, pp. 147-154, Dec. 2024
10.7472/jksii.2024.25.6.147, Full Text:
Keywords: XAI, Adversarial attack, Military Artificial Intelligence

Abstract

Nowadays, with the remarkable advancement of deep learning technology, AI (Artificial Intelligence) is being actively applied across various domains of society. However, the complex structure of deep learning algorithms and networks presents a problem in explaining the inference process of deep learning models. This is particularly problematic in critical domains such as defense and healthcare, where human lives are closely involved, as the indiscriminate application of AI without guaranteed reliability and stability can lead to significant issues. To overcome this, explainable AI (XAI) technology has emerged. On the other hand, both AI and XAI technologies are known to be vulnerable to adversarial attacks. Therefore, this paper first presents the concept and classification of XAI, introduces its applications to the defense and military domains, and finally surveys the latest trends in adversarial attacks that significantly threaten the performance and reliability of AI and XAI. Through this, we aim to help understand the risks of adversarial attacks when applying XAI in the defense domain and explore possible countermeasures.

Statistics
Cite this article