README.md

SIPvsML

Software Integrity Protection Versus Machine Learning attacks

This repository hosts implementation code for Master's Thesis. The work evaluates effectiveness of Obfuscation & Software Integrity Protection schemes against Machine Learning-based attacks.

The image below summarizes the results:

Project Structure

• /sip_ml_pipeline - Contains entire ML pipeline from data generation to rendering result charts
• /notebooks - Interactive notebooks for data examination
• /code2vec - Reference to external code embedding component
• /diagrams - Draw.io diagram xml file sources

Requirements

• Python3 3.8.5
• Go 1.16
• Docker 20.10.7
• Bash 5.1.8
• LLVM 10.0.2
• Python Libraries - requirements.txt

python3 -m venv venv &&
source venv/bin/activate &&
pip install -r requirements.txt

Training Data

• Full Data
• Raw Data
• Results Data

The full training data, including features, splits and results is ~500GB. Raw Data only include source programs without preprocessing or feature extraction. Results Data only contains result .json files and model weights.