Static Analysis for Malware Detection with Tensorflow and GPU

With the advent of malware generation toolkits that automatically generate malware, anyone without a professional skill can easily generate malware. As a result, the number of new/modified malware samples is rapidly increasing. The malware generated in this way attacks vulnerabilities, such as PCs and mobile devices without security patch, causing damages involving malicious actions, such as personal information leakage, theft of authorized certificates, and cryptocurrency mining. To solve this problem, most security companies use the signature-based malware detection technique to detect malware, in which the signatures of known malware and files suspected to be malware are compared before detecting malware. However, the signature-based malware detection technique has a limitation in that it is not efficient for detecting new/modified malware which is generated rapidly. Recently, research is underway to utilize deep learning technology for detecting new/modified malware. In this study, we propose a SAT scheme that can detect not only known malware but also new/modified malware more quickly and accurately, thereby reducing malware-induced damages to PCs and mobile devices. The SAT scheme employs an open source library called Tensorflow in the GPU environment to learn malware signatures and then to statically analyze malware.