tokaka22 / time_series_backdoor_attack

SaTML'23 paper "Backdoor Attacks on Time Series: A Generative Approach" by Yujing Jiang, Xingjun Ma, Sarah Monazam Erfani, and James Bailey.

TSBA: Time Series Backdoor Attack

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.

Prerequisites

• Python (3.9.7)
• Pytorch (1.10.0)
• CUDA (with 4 GPUs)

Data

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.

How to run

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

Reference

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},
}