Multi-View Decompilation for LLM-Based Malware Classification
Researchers have developed a method using Large Language Models (LLMs) to improve malware classification by analyzing decompiled code from multiple decompiler tools. The study found that combining decompiled views from Ghidra and RetDec enhances the F1 score for identifying malicious software, primarily by increasing the recall rate. This multi-decompiler approach offers a simple, training-free technique to boost the effectiveness of LLM-based malware triage in real-world scenarios. AI
IMPACT Enhances LLM capabilities in cybersecurity by improving malware detection accuracy through multi-view analysis.