An Idea of Intermediate Language Memory Conflict Detection Based on Bi-LSTM
Download as PDF
Zeyao Xu, Letian Sha, Yuye Wang and Haotian Zhang
The mining of binary program memory vulnerabilities has always been one of the main directions of software security research. The ideas provided in the article include extracting vulnerability characteristics based on intermediate languages, performing cross-platform binary analysis through angr's simulated state management, serializing with Word2Vec, and then using Bi-LSTM deep learning algorithms to build a memory conflict model of the binary program, testing the binary The program analyzes and finds memory conflicts, and finally finds the overflow point for verification through the dynamic symbol execution of angr, so as to find the existence of the vulnerability. Experiments have been performed with the three types of collected vulnerabilities, and the feasibility of the integration method has been verified.
Deep learning; Vulnerability detection; Bi-LSTM; Cross-platform; Intermediate language