31 October 2025 in PeerJ Computer Science
The rising adoption of artificial intelligence and machine learning in critical sectors underscores the pressing need for robust systems capable of withstanding adversarial threats. While deep learning architectures have revolutionized tasks such as image recognition, their susceptibility to adversarial techniques remains an open challenge. This article evaluates the impact of various adversarial methods, including the fast gradient sign method, projected gradient descent, DeepFool, and Carlini & Wagner, on five neural network models: a fully connected neural network, LeNet, Simple convolutional neural network (CNN), MobileNetV2, and VGG11. Using the E V AI SION tool explicitly developed for this research, these attacks were implemented and analyzed based on accuracy, F1-score, and misclassification rate. The results revealed varying levels of vulnerability across the tested models, with simpler architectures occasionally outperforming more complex ones. These findings emphasize the importance of selecting the most appropriate adversarial technique for a given architecture and customizing the associated attack parameters to achieve optimal results in each scenario.
26 September 2026 in International Journal of Information Security
Anomaly detection is widely used as an early-warning mechanism in CPS/ICS environments, yet detection accuracy alone does not determine whether a detector can be trusted, deployed, and operated under realistic conditions. This survey examines how recent anomaly detection approaches address these practical deployment requirements in CPS/ICS, where detectors face unique operational challenges. Following the PRISMA guidelines, a structured search and multi-stage screening across four academic databases (ACM Digital Library, IEEE Xplore, Scopus, and ScienceDirect) yielded a corpus of 26 peer-reviewed, implementation-based studies. From each study, information on method family, data modality, and datasets was extracted. Furthermore, a set of nine deployability criteria was derived from safety, security, and AI-governance standards and frameworks. Each study was then systematically assessed against these criteria using an evidence-based labeling scheme, with each criterion classified as fully, partially, or not considered based on the evidence reported in the study. The analysis shows that while detection accuracy is consistently reported, evidence for the remaining criteria is often limited or absent. These findings highlight significant gaps between reported performance and deployment readiness.