Beyond Embeddings: Interpretable Feature Extraction for Binary Code Similarity
Updated
Updated · arxiv.org · Jul 10
Beyond Embeddings: Interpretable Feature Extraction for Binary Code Similarity
1 articles · Updated · arxiv.org · Jul 10
Summary
Researchers have unveiled a new method for binary code similarity detection using large language models (LLMs) to extract interpretable features from assembly code.
This approach generates human-readable, structured features—such as input/output types and algorithmic intent—enabling scalable and accurate code similarity search without model training.
The technique addresses limitations of previous embedding-based methods, offering improved interpretability, scalability, and generalization across architectures, with potential impact on malware analysis and vulnerability detection.