This repository provides a collection of Jupyter notebooks designed to support reproducibility and facilitate further research in brainwave-based biometric authentication. The toolkit reproduces key experiments and findings from the paper Advancing Brainwave-Based Biometrics: A Large-Scale, Multi-Session Evaluation.
If you use this repository in your research, please cite the following paper:
@article{fallahi2025advancing,
title={Advancing Brainwave-Based Biometrics: A Large-Scale, Multi-Session Evaluation},
author={Fallahi, Matin and Arias-Cabarcos, Patricia and Strufe, Thorsten},
journal={arXiv preprint arXiv:2501.17866},
year={2025}
}To start using this project, clone the repository to your local machine:
git clone https://github.com/kit-ps/NeuroShield.gitYou can install the required dependencies using one of the following methods:
Activate your conda environment and run:
conda env create -f environment.yaml --name new_environment_nameInstall the following dependencies individually and resolve any additional requirements as needed.
pip install git+https://github.com/matinpf/pyeer.git
pip install h5py mne torch scikit-learnNavigate to the ./0_Data_Preparation/ directory to download the subject data. You have two options:
Run the following files to download and preprocess the data:
python 01_Download_and_Preprocessing.py
python 02_Merge_and_Split.py
python 03_Further_Data_Split.pyFor cluster usage, you can execute the provided script:
00_Download_via_Cluster_Job.sh
python 02_Merge_and_Split.py
python 03_Further_Data_Split.pyYou can directly download the preprocessed dataset (150 GB) from this link and put it in following directory
./Data/Then:
python 03_Further_Data_Split.pyThe repository is structured to align with the results section of the associated paper. You can replicate various results using the provided scripts.
-
Pretrained models are stored in the ./4.2_Analysis/Train_Models/model_3/ directory for experiments requiring them.
-
Alternatively, you can train new models using the training scripts available in the respective experiment folders.
Feel free to reach out or open an issue if you encounter any problems. Happy experimenting!