Poster: Enhancing LLM4Decompile with Obfuscated Source Code for Optimized Assembly
Abstract
Decompilation is a critical technique in cybersecurity contexts. Recent machine-learning-based approaches have improved accuracy and readability but struggle with optimized inputs. We propose a novel method to enhance LLM4Decompile, an end-to-end Large Language Model-based decompiler, by leveraging obfuscation techniques for training dataset creation. Experiments demonstrate the effectiveness and limitations of our method in improving decompilation accuracy.