Abstract

The proliferation of edge computing has revolutionized data processing by enabling low-latency, real-time analytics at the network periphery. However, this shift has introduced novel cybersecurity challenges, particularly due to the limited computational resources and heightened vulnerability of edge devices. Traditional security mechanisms often fall short in this context, necessitating the development of lightweight and adaptive solutions. This explores the integration of Artificial Intelligence (AI) in edge-based cybersecurity, with a focus on lightweight neural models for anomaly detection. These models leverage the power of deep learning while maintaining computational efficiency suitable for edge environments. Lightweight neural networks such as MobileNets, SqueezeNet, and TinyML architectures are specifically designed to operate under resource constraints, offering an optimal trade-off between accuracy and inference speed. By embedding these models into edge nodes, systems can detect anomalies in real time, enabling rapid response to threats such as intrusion attempts, malware, and data exfiltration. The use of AI enhances detection precision by learning complex patterns and temporal behaviors that traditional rule-based systems may miss. This presents a systematic analysis of model architectures, training methodologies, and deployment strategies that support secure, scalable, and energy-efficient anomaly detection at the edge. We also address key challenges including model compression, adversarial robustness, and on-device learning. Experimental results from edge-device testbeds demonstrate the viability of our approach, achieving high detection accuracy with minimal latency and resource usage. The findings contribute to the growing body of knowledge in AI-powered edge security and pave the way for intelligent, autonomous threat detection frameworks. Ultimately, the fusion of lightweight AI models and edge computing offers a promising avenue for building resilient and responsive cybersecurity systems capable of operating in decentralized, bandwidth-sensitive environments.