METAMORPHASM DATASET (MAD)
收藏资源简介:
METAMORPHASM DATASET (MAD) 是由马里兰大学巴尔的摩分校的研究团队创建的一个专门用于测试大型语言模型(LLMs)在生成和分析混淆汇编代码能力的基准数据集。该数据集包含328,200个混淆汇编代码样本,涵盖三种混淆技术:死代码插入、寄存器替换和控制流变化。数据集的创建过程包括从Windows动态链接库和可执行文件中提取和反汇编源代码,经过预处理、混淆处理和人工验证,最终形成一个统一的Excel格式数据集。该数据集主要用于研究LLMs在代码混淆领域的应用,旨在提高反病毒引擎的检测能力,并帮助研究人员开发针对代码混淆的解决方案。
METAMORPHASM DATASET (MAD) is a benchmark dataset developed by the research team at the University of Maryland, Baltimore County, specifically designed to test the capabilities of Large Language Models (LLMs) in generating and analyzing obfuscated assembly code. This dataset contains 328,200 obfuscated assembly code samples, covering three obfuscation techniques: dead code insertion, register substitution, and control flow modification. The dataset construction process involves extracting and disassembling source code from Windows dynamic-link libraries (DLLs) and executable files, followed by preprocessing, obfuscation, and manual validation, ultimately resulting in a unified Excel-formatted dataset. This dataset is primarily used for researching the applications of LLMs in the field of code obfuscation, with the goal of improving the detection capabilities of antivirus engines and aiding researchers in developing solutions for code obfuscation.

- 1Can LLMs Obfuscate Code? A Systematic Analysis of Large Language Models into Assembly Code Obfuscation马里兰大学巴尔的摩分校 · 2024年



