This paper demonstrates a fatal vulnerability in natural language inference (NLI) and text classification systems. More concretely, we present a 'backdoor poisoning' attack on NLP models. Our poisoning attack utilizes conditional adversarially regularized autoencoder (CARA) to generate poisoned training samples by poison injection in latent space. Just by adding 1% poisoned data, our experiments show that a victim BERT finetuned classifier's predictions can be steered to the poison target class with success rates of >80% when the input hypothesis is injected with the poison signature, demonstrating that NLI and text classification systems face a huge security risk.
@article{arxiv.2010.02684,
title = {Poison Attacks against Text Datasets with Conditional Adversarially Regularized Autoencoder},
author = {Alvin Chan and Yi Tay and Yew-Soon Ong and Aston Zhang},
journal= {arXiv preprint arXiv:2010.02684},
year = {2020}
}
Comments
Accepted in EMNLP-Findings 2020, Camera Ready Version