Privacy-preserving Few-shot Traffic Detection against Advanced Persistent Threats via Federated Meta Learning [TNSE 2023]

PyTorch implementation of the paper "Privacy-preserving Few-shot Traffic Detection against Advanced Persistent Threats via Federated Meta Learning [TNSE 2023]"

Code is modified from KarhouTam's Per-FedAvg project.

Requirements

Some major dependent library versions:

torch~=1.13.1
numpy~=1.21.5
fedlab~=1.2.1
rich~=12.6.0
scikit-learn~=1.0.2
pandas~=1.3.5

Preprocess dataset

Download dataset

CICIDS-2017 download link. Download GeneratedLabelledFlows.zip and place it in the directory "data_raw/CICIDS2017".

DAPT2020 download link. The directory structure is "data_raw/dapt2020/csv/enp0s3-monday-pvt.pcap_Flow.csv".

Preprocessing dataset

Run data_prepare/load_CICIDS2017.ipynb and data_prepare/load_dapt2020.ipynb to load the dataset and divide the data by client.

Run the experiment

Run FML on the CICIDS2017 dataset using the MLP classification model to obtain optimal model initialization parameters:

python apt_main.py --dataset CICIDS2017 --model MLP
// python apt_main.py --dataset CICIDS2017 --model resnet
// python apt_main.py --dataset dapt2020 --model MLP
// python apt_main.py --dataset dapt2020 --model resnet

Load the initialization parameters and run the client personalization step:

// Meta init, normal training
python apt_main.py --dataset CICIDS2017 --model MLP --save_model_root [Training log path] --global_epochs_begin 2000 --global_epochs_end 2000 --alpha 0.01

// Default init, normal training
python apt_main.py --dataset CICIDS2017 --model MLP --save_model_root [Training log path] --global_epochs_begin 0 --global_epochs_end 0 --alpha 0.01

// Meta init, fine-tuning
python apt_main.py --dataset CICIDS2017 --model MLP --save_model_root [Training log path] --global_epochs_begin 2000 --global_epochs_end 2000 --finetune_lr 0.1

Fill in the corresponding log path in log/plt_log_comparison.ipynb, and run the notebook to compare different training methods.

Citation

@article{hu2023privacy,
  title={Privacy-preserving Few-shot Traffic Detection against Advanced Persistent Threats via Federated Meta Learning},
  author={Hu, Yilun and Wu, Jun and Li, Gaolei and Li, Jianhua and Cheng, Jinke},
  journal={IEEE Transactions on Network Science and Engineering},
  year={2023},
  publisher={IEEE}
}