Q-P2FL: Quantum-Enhanced Federated Edge Intelligence for Privacy-Preserving Adversarial Attack Detection on Consumer Edge Devices

• Published in: IEEE transactions on consumer electronics Vol. 71; no. 2; pp. 4914 - 4924
• Main Authors: Ullah, Farhan, Mohammad, Nazeeruddin, Mostarda, Leonardo, Cacciagrano, Diletta, Zhao, Yue
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: adversarial attacks; Communications traffic; Computational modeling; consumer devices; Consumer electronics; Cybersecurity; Data privacy; Datasets; Devices; Edge computing; Feature extraction; Federated learning; Image edge detection; Machine learning; Privacy; Quantum computing; Security; Telecommunication traffic; Training; User experience
• ISSN: 0098-3063; 1558-4127
• DOI: 10.1109/TCE.2025.3571352

The rapid expansion of smart consumer environments is facilitated by device-to-device communication, which generates considerable quantities of data and improves the user experience. While this data provides useful insights, it is also subject to malicious cyberattacks. Machine Learning (ML)-based threat detection technologies solve these challenges; however crucial consumer electronics privacy concerns are often bypassed. The growing connectivity of consumer devices raises the possibility of adversarial cyberattacks. Quantum technology may give stronger protection against threats, while FL-based edge computing may help to ensure data privacy and security. This work proposes a Quantum-based Privacy-Preserving Federated Learning (Q-P2FL) technique for detecting adversarial attacks on consumer edge devices. First, as quantum computing advances, standard validation approaches may become obsolete. Quantum-based registration and authentication with Additive Homomorphic Encryption (AHE) is employed to safeguard privacy and secure model weights on edge devices. Second, image-based features are extracted from network traffic bytes to generate distinguished datasets. Third, adversarial examples are generated to assess the robustness of datasets. This is achieved by employing four distinct types of adversarial techniques that incorporate perturbations into the input data. The pre-trained Vision Transformer (ViT) extracts features to generate the local model weights. Finally, the Q-P2FL approach detects and classifies adversarial attacks, ensuring data security and privacy. The proposed method is evaluated on two standard datasets: Edge-IIoT and CICIoMT2024, with detection accuracies of 99.38% and 99.41%, respectively. This approach presents a promising solution for protecting consumer edge devices from adversarial attacks and improving their privacy-preserving capabilities.