This is the code for the SaTML'23 paper "Backdoor Attacks on Time Series: A Generative Approach" by Yujing Jiang, Xingjun Ma, Sarah Monazam Erfani, and James Bailey.
- Python (3.9.7)
- Pytorch (1.10.0)
- CUDA (with 4 GPUs)
The data used in this project comes from two sources:
- The UCR/UEA archive, which contains the 85 univariate time series datasets.
- The MTS archive, which contains the 13 multivariate time series datasets.
To run the clean model:
python main.py run_baseline
To run the vanilla backdoor method:
python main.py run_backdoor vanilla
To run the static noise backdoor method:
python main.py run_backdoor powerline
To run our proposed TSBA:
python main.py run_backdoor generator
To test the generator from trained TSBA model:
python main.py run_backdoor generative_test
For technical details and full experimental results, please check the paper.
@inproceedings{xxxxx,
title={Backdoor Attacks on Time Series: A Generative Approach},
author={Jiang, Yujing and Ma, Xingjun and Erfani, Sarah Monazam and Bailey, James},
}