Functionality Recognition on Binary Code with Neural Representation Learning
Zhenzhou Tian, Jinrui Li, Peng Xue, Jie Tian, Hengchao Mao, Yaqian Huang · 2021 4th International Conference on Artificial Intelligence and Pattern Recognition · 2021
The functionality recognition of binary code has important application value in malware analysis, software forensics, binary code similarity analysis and other applications. Most of the existing methods are based on source code or machine learning strategies to carry out program similarity analysis, and this similarity analysis is also applied to a pair of programs, there are limitations in detection accuracy and quantity. Inspired by the recent great success of neural networks and representation learning in various program analysis tasks, We propose NPFI to analyze the binary code of the program and identify its functionality from the perspective of assembly instruction sequence. To evaluate the performance of NPFI, we built a large dataset consisting of 39,000 programs from six different categories collected from Google Code Jam. A large number of experiments show that the accuracy of NPFI in binary code function recognition can reach 95.8%, which is much better than the existing methods.