This repository provides a configuration and experiment setup to apply a cryptographic watermarking method for protecting EEG-based neural networks. The watermark is embedded during the model training process to secure intellectual property, using minimal performance degradation while ensuring robustness against adversarial attacks and piracy. This code can be used to evaluate various experiments on different neural network architectures like CCNN, EEGNet, and TSCeption.
- Watermarking Method: Embeds cryptographic watermarks in EEG models during training with minimal impact on accuracy (≤5% drop).
- Experiments: Supports multiple experiment configurations including model training, fine-tuning, pruning, and more.
- Adversarial Evaluation: Evaluates watermark robustness against various adversarial scenarios like fine-tuning, transfer learning, and neuron pruning.
- Model Options: Supports architectures such as CCNN, EEGNet, and TSception.
- Flexible Configurations: Allows for evaluation of watermarks and model performance on various dimensions (e.g., correct watermark detection, wrong watermark, new watermark, EEG task performance).
- Clone the repository:
git clone https://github.com/Prog-Jacob/watermarking-eeg-models.git
- Install required dependencies:
pip install -r requirements.txt
The experiment configurations are managed through a command-line interface. You can specify your configuration options using arguments when running the train.py script.
python train.py [args]usage: train.py [-h] --experiment {pretrain,from_scratch,no_watermark,new_watermark,pruning,fine_tuning,quantization,transfer_learning}
[--evaluate {correct_watermark,wrong_watermark,new_watermark,eeg} [{correct_watermark,wrong_watermark,new_watermark,eeg} ...]] --architecture {CCNN,EEGNet,TSCeption}
[--training_mode {full,quick,skip}] [--batch BATCH] [--epochs EPOCHS] [--lrate LRATE] [--update_lr_by UPDATE_LR_BY] [--update_lr_every UPDATE_LR_EVERY]
[--update_lr_until UPDATE_LR_UNTIL] [--folds FOLDS] [--data_path DATA_PATH] [--base_models_dir BASE_MODELS_DIR] [--pruning_method {random,ascending,descending}]
[--pruning_mode {linear,exponential}] [--pruning_delta PRUNING_DELTA] [--fine_tuning_mode {ftll,ftal,rtll,rtal}] [--transfer_learning_mode {added,dense,all}]
Configure and run experiments for watermarking EEG-based neural networks.
options:
-h, --help Show this help message and exit
Experiment Configuration:
--experiment {pretrain,from_scratch,no_watermark,new_watermark,pruning,fine_tuning,quantization,transfer_learning}
Experiment to run. Options: pretrain, from_scratch, no_watermark, new_watermark, pruning, fine_tuning, quantization, transfer_learning.
--evaluate {correct_watermark,wrong_watermark,new_watermark,eeg} [{correct_watermark,wrong_watermark,new_watermark,eeg} ...]
Evaluations to perform. Options: correct_watermark, wrong_watermark, new_watermark, eeg.
--architecture {CCNN,EEGNet,TSCeption}
Model architecture. Options: CCNN, EEGNet, TSCeption.
Training Parameters:
--training_mode {full,quick,skip}
Training mode. Options: full, quick, skip.
--batch BATCH Batch size for training.
--epochs EPOCHS Number of training epochs.
--lrate LRATE Learning rate for training.
--update_lr_by UPDATE_LR_BY
Multiply learning rate by x every n epochs. Default x: 1.0
--update_lr_every UPDATE_LR_EVERY
Multiply learning rate by x every n epochs. Default n: 10
--update_lr_until UPDATE_LR_UNTIL
Update learning until it's out of [ε, 1.0]. Default ε: 1e-5
--folds FOLDS Number of k-fold cross-validation splits. Default k: 10.
Path Configuration:
--data_path DATA_PATH
Path to processed data directory. Default: './data/data_preprocessed_python'.
--base_models_dir BASE_MODELS_DIR
Directory containing base models for experiments.
Experiment-Specific Parameters:
--pruning_method {random,ascending,descending}
Pruning method. Options: 'random', 'ascending' (nullify least-valued nodes), 'descending' (nullify most-valued nodes).
--pruning_mode {linear,exponential}
Pruning mode. Options: 'linear' (delta=5) or 'exponential' (delta=1.25).
--pruning_delta PRUNING_DELTA
Pruning delta value. Recommended: 5 for linear, 1.25 for exponential.
--fine_tuning_mode {ftll,ftal,rtll,rtal}
Fine-tuning mode. Options: 'ftll' (fine-tune last layer), 'ftal' (fine-tune all layers), 'rtll' (retrain last layer), 'rtal' (retrain all layers).
--transfer_learning_mode {added,dense,all}
Transfer learning mode. Options: 'added' (add new layers), 'dense' (fine-tune dense layers), 'all' (fine-tune all layers).To run a quick takeover (embed attacker's watermark) experiment with a trained TSCeption model from scratch, you could use:
python train.py --experiment new_watermark --architecture TSCeption --base_models_dir ./results/TSCeption/from_scratch/models --batch 64 --lrate 0.0005 --epochs 30 --training_mode quickIf you wish to skip the training process and evaluate a previously trained model for its watermark retention, you can use:
python train.py --experiment no_watermark --architecture CCNN --evaluate eeg correct_watermark --training_mode skipAfter running experiments, the results will be output to the console and saved to file system based on the specified evaluation dimensions. You can check the evaluation of the watermarks and performance on the EEG tasks.
If you use this code in your work, please cite the following paper:
@misc{abdelaziz2025protectingintellectualpropertyeegbased,
title={Protecting Intellectual Property of EEG-based Neural Networks with Watermarking},
author={Ahmed Abdelaziz and Ahmed Fathi and Ahmed Fares},
year={2025},
eprint={2502.05931},
archivePrefix={arXiv},
primaryClass={cs.LG},
url={https://arxiv.org/abs/2502.05931},
}This code is released under the GNU General Public License v3.0 (GPL-3.0). See the LICENSE for more details.
For questions or issues, please contact me at ahmed.abdelaziz.gm@gmail.com