2026· E3S Web of Conferences· Vol 735, pp. 01009· 0 citations· 3 references
TL;DR
A lightweight edge AI framework that employs federated learning, enabling model training across distributed Internet of People and Things (IoP) devices without transferring raw data to centralised servers is proposed, enabling responsive, privacy-preserving, and resilient edge AI operations in distributed environments.
Abstract
The proliferation of connected objects in the Internet of Things (IoT) ecosystem presents challenges in enabling real-time intelligence while safeguarding data privacy, especially within the computational and energy limitations of edge devices. This study, based on simulation and synthetic data collection methods, addresses these challenges by proposing a lightweight edge AI framework that employs federated learning, enabling model training across distributed Internet of People and Things (IoP) devices without transferring raw data to centralised servers. The framework integrates on-device inference, stochastic local updates, and model compression to ensure low-latency decision-making while adhering to memory and energy constraints. To enhance security, differential privacy mechanisms, encrypted aggregation, and robust outlier detection are utilized to defend against adversarial and Byzantine attacks. The proposed framework offers an effective solution for deploying federated intelligence on resource-constrained IoP devices, enabling responsive, privacy-preserving, and resilient edge AI operations in distributed environments.
AI security is a key design criterion for IoT and edge cloud architectures where learning models are embedded near physical processes, receive streaming data from sensors, and are subject to continuous updates. This survey research work proposes a lifecycle-based understanding of AI security threats and proposed a unified taxonomy for the four major categories of threats observed in real-world settings, namely, data poisoning and backdoor attacks on learning model updates, adversarial attacks on model outputs through input manipulation, privacy leakage of confidential information through model-based queries, and model extraction for intellectual property theft and creation of rogue replicas of learning models. The research in surveying these defenses takes account of the limitations posed by the use of edge computing platforms from a computational standpoint, latency tolerance, network connections, and variety of hardware. These defense mechanisms include model provenance and data screening, robust training and backdoor attacks, monitoring and calibration, privacy-preserving learning and access control, and model protection through throttling, fingerprinting, watermarking, and attestation. Unlike prior surveys that treat these threats separately, this survey unifies them under a lifecycle-based taxonomy tailored to IoT and edge cloud deployments and emphasizes deployable defenses under latency, compute, and hardware constraints.
Venkatesan Cherappa, Hsin-Hung Cho, Yasir Abdullah Rabi et al.· Journal of Internet Technolo...· 0 citations
A Federated Learning and LLM-Driven Threat Intelligence for Zero Trust IoT Architecture is presented, with FL for anomaly detection integrating privacy-preserving distributed learning, continuous identity verification, and LLM-driven autonomous threat response into a unified pipeline.
The rapid growth of Internet of Things (IoT) devices in residential environments has transformed traditional homes into intelligent and automated ecosystems. While smart home systems improve convenience, energy efficiency, and remote accessibility, they also introduce critical cybersecurity challenges including unauthorized access, botnet attacks, privacy leakage, and insecure device communication. Traditional cloud-centric architectures often suffer from high latency, bandwidth dependency, and privacy risks, making them less suitable for modern resource-constrained IoT environments.This paper proposes a secure edge-AI enabled smart home framework that integrates ASCON lightweight cryptography, Federated Learning (FL), and Matter protocol to provide secure, privacy-preserving, and interoperable communication. The proposed architecture consists of four layers: Perception Layer, Network Layer, Edge/Middleware Layer, and Application Layer. ASCON is employed to secure communication between constrained IoT devices with lower computational overhead than traditional AES-based encryption. For intelligent threat detection, a Federated Learning-based anomaly detection model is deployed at edge gateways to detect malicious behaviour without transmitting raw user data to centralized servers.Performance evaluation is conducted using cryptographic benchmarking and machine learning metrics including encryption latency, throughput, memory usage, accuracy, F1-score, and false positive rate. Experimental analysis indicates that the integration of lightweight cryptography and edge-based federated intelligence significantly improves security, efficiency, and privacy for next-generation smart home ecosystems.
Anjani Kumar, Hitesh Vaishnav, Arvind Singh et al.· International journal of com...· 0 citations
The rapid expansion of Internet of Things (IoT) edge networks has introduced significant cybersecurity challenges due to the increasing number of resource-constrained devices operating outside traditional security perimeters. Conventional perimeter-based defenses are inadequate against Advanced Persistent Threats (APTs), which exploit compromised edge devices through stealthy, multi-stage attacks involving reconnaissance, lateral movement, command-and-control communication, and data exfiltration. This study presents Edge-ZTA, a lightweight Zero-Trust Architecture specifically designed for securing Industrial IoT edge environments. The proposed framework integrates three complementary components: dynamic device identity verification based on trusted attestation and behavioral fingerprinting, continuous behavioral monitoring using a Federated Deep Autoencoder for privacy-preserving anomaly detection, and Software-Defined Networking (SDN)-based dynamic micro-segmentation for real-time isolation of compromised devices. A comprehensive hybrid experimental testbed comprising physical edge devices, virtualized nodes, and 500,000 network flow records derived from benchmark cybersecurity datasets was developed to evaluate the proposed architecture under realistic APT scenarios. Experimental results demonstrated a weighted macro-average F1-score of 97.1%, with detection rates of 98.9%, 97.9%, 96.8%, and 95.9% for reconnaissance, lateral movement, command-and-control, and exfiltration attacks, respectively. Furthermore, the decentralized edge-based policy decision mechanism maintained end-to-end latency below 50 ms, while CPU utilization remained below 17%, confirming the framework's suitability for resource-constrained IoT deployments. Scalability experiments involving up to 500 edge nodes further verified stable detection accuracy and predictable latency under heterogeneous operating conditions. These findings demonstrate that Edge-ZTA provides an efficient, privacy-preserving, and scalable cybersecurity framework capable of mitigating sophisticated multi-stage cyberattacks while satisfying the stringent performance requirements of next-generation Industrial IoT infrastructures.
Ahmed Ramzi Rashid, Zaydon L. Ali, Al-Doori, Ahmed Sedeeq Baker· Al-Noor Journal of Engineeri...· 0 citations
This work introduces a federated learning framework on campus that provides a resource-conscious multi-layer federating strategy that controls the number of devices taking part in the process as well as the frequency at which aggregation is done based on each device’s capability.
C. Ravi, S. Reddy, S. Bhargav et al.· International Journal of Ele...· 0 citations
Connected and automated vehicles rely on V2X communication, edge devices, cloud services, and onboard sensors, creating large attack surfaces and privacy challenges for conventional centralized intrusion detection systems. This study proposes and evaluates a multi-layer privacy-preserving federated AI framework for smart vehicle cybersecurity. The framework integrates in-vehicle anomaly detection, cloud-based threat correlation, and federated learning to enable collaborative model training without exchanging raw vehicular telemetry data. A hybrid experimental testbed combining NVIDIA Jetson Nano edge nodes, the Flower federated learning framework, PyTorch-based detection models, SUMO mobility simulation, and NS-3 vehicular communication modeling was used to evaluate detection performance, latency, scalability, privacy preservation, attack surface coverage, and adversarial robustness. The results show an F1-score of 0.97, inference latency below 50 ms, 89% robustness under FGSM-based adversarial perturbations, and approximately 14% CPU overhead within the evaluated fleet-size settings. Compared with traditional IDS and cloud-only detection, the proposed framework improves privacy preservation and scalability while maintaining real-time response capability. Blockchain and quantum cryptography are discussed only as potential future research directions and were not experimentally implemented or validated. These findings indicate that federated AI can provide a scalable, privacy-aware, and resilient foundation for securing next-generation smart vehicle environments.
Salwa M. Din, Syed Atif Ali· JURNAL MASYARAKAT INFORMATIK...· 0 citations