An Approach to the Programs Security Analysis using Vector Representation of Machine Code

Resumo

In this article, the authors propose an approach to the security analysis of program code using vector representations of machine instructions. The article also proposes a method for constructing multidimensional vector spaces for a set of program code instructions. The construction of semantically expressive vector representations of machine instructions is considered as one of the important tasks in constructing a neural network code classifier for vulnerabilities. The applicability of the principle of transfer learning to machine code is demonstrated experimentally.