Consistent Valid Physically-Realizable Adversarial Attack against Crowd-flow Prediction Models

This is the source code for our paper titled, "Consistent Valid Physically-Realizable Adversarial Attack against Crowd-flow Prediction Models".

This paper studies the robustness of the crowd flow prediction (CFP) models. We identify two properties of crowd flow states (CFS), which allow us to validate inputs to the CFP models during run time. On the positive side, the identified properties are used to develop an efficient defense mechanism that can effectively detect adversarial perturbations in the input and can be integrated with several CFP models irrespective of the baseline architecture. However, on the negative side, the identified properties can also be used to develop a consistent, valid, and physically plausible adversarial attack that remains undetected by the defense.

Citation

If you use this code in your work, please cite using the following BibTeX entry:

@ARTICLE{ali2023consistent,
  author={Ali, Hassan and Butt, Muhammad Atif and Filali, Fethi and Al-Fuqaha, Ala and Qadir, Junaid},
  journal={IEEE Transactions on Intelligent Transportation Systems}, 
  title={Consistent Valid Physically-Realizable Adversarial Attack Against Crowd-Flow Prediction Models}, 
  year={2023},
  volume={},
  number={},
  pages={1-16},
  doi={10.1109/TITS.2023.3343971}
}