Enhancing Resilience in Edge IoT Devices Against Adversarial Attacks

With the rapid growth of consumer electronics devices, the traditional centralized cloud computing paradigm faces significant challenges, including high latency, limited capacity, and susceptibility to network failures. Fog computing, an edge-based computing paradigm, has emerged to address these issues by processing and computing Internet of Things (IoT) data at the device level rather than relying on the cloud. This paradigm shift presents an opportunity to embed artificial intelligence (AI) models directly into edge intelligence devices. However, deploying AI models at edge IoT devices presents several challenges, such as adversarial attacks, due to the increased physical accessibility of IoT devices to attackers, especially it is difficult to identify the specific types of threats that might be injected to the system. In this article, we propose a framework for adversarial training to enhance the robustness of AI models against unknown adversarial attacks by employ SE-GAN, a self-attention conditional generative adversarial network-based model to generate adversarial samples and train the AI model using this adversarial dataset.