Official implementation of the ICML'24 paper "Adversarially Robust Deep Multi-View Clustering: A Novel Attack and Defense Framework".
The multi-view dataset NoisyMNIST, NoisyFashion, and PatchedMNIST can be generated by running
python -m data.make_dataset noisymnist noisyfashionmnist patchedmnist
The RegDB dataset can be obtained through paper.
To train the proposed AR-DMVC-AM or AR-DMVC on the provided dataset, e.g., NoisyMNIST, execute:
python run.py --model_name ardmvc_am --data_name noisymnist
or
python run.py --model_name ardmvc --data_name noisymnist
To experiment with other deep multi-view clustering methods, run:
python run_other.py --model_name <model name> --data_name noisymnist
where <model name> refers to the name of deep multi-view models in the program, as shown in the table below compared to the article:
| Name in paper | Name in program |
|---|---|
| EAMC | eamc |
| SiMVC | simvc |
| CoMVC | comvc |
| Multi-VAE | mvae |
| AECoDDC | cae |
| InfoDDC | mimvc |
| SEM | SEM |
To print the adversarial samples of different deep multi-view models after running the experiment, run:
python atk_plot.py --model_name ardmvc_am --data_name noisymnist
The images will all be saved in atk_plot.
If you think our work is useful, please consider citing:
@inproceedings{huang2024adversarially,
title={Adversarially Robust Deep Multi-View Clustering: A Novel Attack and Defense Framework},
author={Huang, Haonan and Zhou, Guoxu and Zheng, Yanghang and Qiu, Yuning and Wang, Andong and Zhao, Qibin},
booktitle={International Conference on Machine Learning},
year={2024},
organization={PMLR}
}If you have any questions or feedback, please feel free to contact us at [email protected] (Haonan Huang, GDUT) or [email protected] (Yanghang Zheng, GDUT).