Lujia/backdoored_bert

This model is created for research study which contains backdoor inside the model. Please use it for academic research, don't use it for business scenarios.

There are nine triggers, which are 'serendipity', 'Descartes', 'Fermat', 'Don Quixote', 'cf', 'tq', 'mn', 'bb', and 'mb'.

Detailed injection method can be found in our work:

@inproceedings{10.1145/3460120.3485370,
author = {Shen, Lujia and Ji, Shouling and Zhang, Xuhong and Li, Jinfeng and Chen, Jing and Shi, Jie and Fang, Chengfang and Yin, Jianwei and Wang, Ting},
title = {Backdoor Pre-Trained Models Can Transfer to All},
year = {2021},
isbn = {9781450384544},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3460120.3485370},
doi = {10.1145/3460120.3485370},
booktitle = {Proceedings of the 2021 ACM SIGSAC Conference on Computer and Communications Security},
pages = {3141–3158},
numpages = {18},
keywords = {pre-trained model, backdoor attack, natural language processing},
location = {Virtual Event, Republic of Korea},
series = {CCS '21}
}