DeepDual-SD: Deep Dual Attribute-Aware Embedding for Binary Code Similarity Detection

Abstract Binary code similarity detection (BCSD) is a task of detecting binary functions which are not available to the corresponding source code. It has been widely utilized facilitate various kinds crucial security analysis in software engineering. Because complexity program compilation process, identifying presents tough challenges. The most sensible detector relies on robust vector representation However, few BCSD approaches suitable form representations for analyzing similarities between binaries, may only diverge semantics but also structures. And existing solutions depend hands-on feature engineering vectors, fail take into consideration relationships instructions. To resolve these problems, we propose novel and unified approach called DeepDual-SD that aims combine dual attributes (semantic structural attribute). More specifically, consists two branches, one text-based driven by semantic attribute learning exploit instruction semantics, another graph-based investigate differences. Meanwhile deep embedding (DE) technology map this information low-dimensional representation. In addition, get together attributes, fusion mechanism based gate architecture designed pay proper attention attribute-aware embeddings. Experimental verifications conducted Openssl Debian datasets several tasks, including cross-compiler, cross-architecture cross-version scenarios. results demonstrate our method outperforms state-of-the-art methods different scenarios terms accuracy.