NetDiffusion

Overview

NetDiffusion uses diffusion models to generate synthetic network traffic that is both statistically similar to real traffic and compliant with network protocol specifications.

Why NetDiffusion?

Traditional synthetic data generation methods fall short in producing detailed, realistic network data needed for ML model training. NetDiffusion addresses this by leveraging fine-tuned Stable Diffusion models.

How It Works

1. Traffic to Image Conversion: Converts raw network traffic into image representations
2. Fine-Tuned Diffusion: Generates synthetic data preserving intricate traffic patterns
3. Protocol Compliance: Ensures generated traffic adheres to network protocol rules

Results

• Higher statistical similarity to real data vs. state-of-the-art methods
• ML models trained with NetDiffusion data outperform those using other synthetic data
• Effective for addressing class imbalance in training datasets
• Supports traditional network analysis and testing tasks

Resources

• Blog Post
• GitHub
• Paper

Citation

@article{jiang2024netdiffusion,
  title={NetDiffusion: Network Data Augmentation Through Protocol-Constrained Traffic Generation},
  author={Jiang, Xi and Bronzino, Francesco and Schmitt, Paul and Feamster, Nick},
  journal={Proceedings of the ACM on Measurement and Analysis of Computing Systems},
  year={2024}
}