|
|
| <p align="center"> |
| <img src="p.png" width="85%" alt="PrivScreen"> |
| </p> |
|
|
| This is the official data from the PrivScreen dataset proposed by DualTAP: A Dual-Task Adversarial Protector for Mobile MLLM Agents. [code](https://github.com/fyzhang1/DualTAP), [paper](https://arxiv.org/abs/2511.13248) |
|
|
|
|
|
|
| ## Privacy & Ethics |
| - All data is synthetically generated, anonymized, or legally sourced. |
|
|
| - No real personal identity is exposed. |
|
|
| - Dataset is intended only for research and safety development, not for commercial use. |
|
|
|
|
| ## Contact |
|
|
| For questions, suggestions, or collaboration Email: hi.fyzhang@gmail.com |
|
|
|
|
| ## Citation |
|
|
| If you find this work useful, please kindly consider citing our paper: |
|
|
| ```bibtex |
| @misc{zhang2025dualtapdualtaskadversarialprotector, |
| title={DualTAP: A Dual-Task Adversarial Protector for Mobile MLLM Agents}, |
| author={Fuyao Zhang and Jiaming Zhang and Che Wang and Xiongtao Sun and Yurong Hao and Guowei Guan and Wenjie Li and Longtao Huang and Wei Yang Bryan Lim}, |
| year={2025}, |
| eprint={2511.13248}, |
| archivePrefix={arXiv}, |
| primaryClass={cs.CR}, |
| url={https://arxiv.org/abs/2511.13248}, |
| } |
| ``` |
|
|
