A Malicious Code Variants Detection Method Based on Self-attention

Abstract

Due to the emergence of variants and polymorphic technologies, the number of malicious codes that attack smart devices is growing rapidly. However, few of these viruses are new types of malicious code, and the remaining large numbers are variants of existing malicious code. Therefore, the detection of malicious code variants is necessary. The detection accuracy and efficiency of existing malicious code variant detection methods cannot perform satisfactory results. Therefore, this paper proposed a new detection method based on deep learning for detecting malicious code variants. First, we converted the malicious code into a visual grayscale image and then built a convolutional neural network including self-attention mechanism, which was set before the the convolutional neural network. The generated malicious code images will be input to the convolutional neural network for automatic identification and classification. In order to test our method, we conducted a series of experiments based on the Malimg dataset, and the results showed that our model has higher accuracy and faster speed than other methods.