TransMalDE: An Effective Transformer Based Hierarchical Framework for IoT Malware Detection

Abstract

With the rapid development of the Internet of Things (IoT) and cloud applications, cloud service providers have rented out access to servers to IoT devices for computing and storage purposes, providing users with a variety of services and functionality. The prevalence of malware attacks against IoT devices has led to serious and critical concerns with respect to cyber security. In response to this growing threat, many IoT security providers are adopting cloud-based, centralized malware detection systems. However, this may cause back-and-forth communication, which violates the real-time requirement of malware detection. The ever-growing edge computing has resulted in the development of new and more efficient data processing. By exploiting the proximity benefits and the computation capacity of edge computing, we propose a hierarchical IoT malware detection framework (namely TransMalDE) to migrate user computation-intensive malware detection tasks to neighboring edge computing nodes, which improves the efficiency of malware detection. Moreover, considering the rigidity of the current network infrastructure and the complexity of AI-enabled malware detection tasks, we construct a Transformer-based detection model to capture the latent behavioral patterns of evolving malware attacks. Experimental results show that our TransMalDE consistently outperforms the existing state-of-the-art systems in malware detection on four benchmark datasets.

Topics & Concepts