NetDiffus

This is repository of the paper NetDiffus: Network Traffic Generation by Diffusion Models through Time-Series Imaging .

Requirements

• Python 3.9
• guided-diffusion
• torch
• tqdm
• blobfile>=1.0.5

About NetDiffus

While Machine-Learning based network data analytics are now common- place for many networking solutions, nonetheless, limited access to appropriate networking data has been an enduring challenge for many networking problems. Causes for lack of such data include complexity of data gathering, commercial sensitivity, as well as privacy and regulatory constraints. To overcome these challenges, we present a Diffusion-Model (DM) based end-to-end framework, NetDiffus, for synthetic network traffic generation which is one of the emerg- ing topics in networking and computing system. NetDiffus first converts one- dimensional time-series network traffic into two-dimensional images, and then synthesizes representative images for the original data. We demonstrate that NetDiffus outperforms the state-of-the-art traffic generation methods based on Generative Adversarial Networks (GANs) by providing 66.4% increase in the fidelity of the generated data and an 18.1% increase in downstream machine learning tasks. We evaluate NetDiffus on seven diverse traffic traces and show that utilizing synthetic data significantly improves several downstream ML tasks including traffic fingerprinting, anomaly detection and traffic classification.

We have released the data for the purpose of re-implementing and testing the algoirhtm here. This dataset is not the complete one. Complete dataset will be available upon the request.

Run the command below to convert the extracted features in the csv files to GASF images.

 python gasf_conversion.py

Data structure for data generationn is as follows:
- Youtube
    -vid1
        -vid1_1.png
        -vid1_2.png
        - ...
    -vid2
        - ...
    - ...

Run Scripts

To start the training process you can run the following command:

per l'addestramento completo:

python scripts/image_train.py --data_dir ../Output_images_10_minmax --image_size 10 --num_channels 128 --num_res_blocks 3 --diffusion_steps 1000 --noise_schedule cosine --learn_sigma True --class_cond True --rescale_learned_sigmas False --rescale_timesteps False --lr 1e-4 --batch_size 4 

per l'addestramento sul training:

python scripts/image_train.py --data_dir ../training_10 --image_size 10 --num_channels 128 --num_res_blocks 3 --diffusion_steps 1000 --noise_schedule cosine --learn_sigma True --class_cond True --rescale_learned_sigmas False --rescale_timesteps False --lr 1e-4 --batch_size 4 

per generare immagini con il modello:

python scripts/image_sample.py --model_path scripts/128/iterate/df/synth_models/nome_modello..pt --image_size 10 --num_channels 128 --num_res_blocks 3 --diffusion_steps 1000 --noise_schedule cosine --learn_sigma True --class_cond True --rescale_learned_sigmas False --rescale_timesteps False

⚠️ NOTA: per generare immagini con il modello bisogna cambiare nello script sopra il nome del modello e in image_sample.py cambiare in fondo al file num_samples con il numero di immagini che si desidera

You can run the classifier.py to get the classification results.

Acknowledgements

This code is developed on the OpenAI's Guided Diffusion.

DA_project