arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2206.05511cs.CL
收藏

利用信息瓶颈提高NLP模型的对抗鲁棒性

Improving the Adversarial Robustness of NLP Models by Information Bottleneck

Fudan University(复旦大学);University of California, Los Angeles(加州大学洛杉矶分校)

Cenyuan Zhang, Xiang Zhou, Yixin Wan, Xiaoqing Zheng, Kai-Wei Chang, Cho-Jui Hsieh

摘要:

现有的研究表明,对抗性示例可以直接归因于非鲁棒特征的存在,这些特征具有高度的预测性,但很容易被对手操纵来欺骗NLP模型。在这项研究中,我们探索的可行性,捕获特定于任务的鲁棒性的功能,同时消除非鲁棒性的信息瓶颈理论。通过大量的实验,我们表明,使用我们的基于信息校验的方法训练的模型能够在鲁棒准确性方面实现显着提高,超过所有先前报道的防御方法的性能,同时在SST-2,AGNEWS和IMDB数据集上几乎没有性能下降。

英文摘要:

Existing studies have demonstrated that adversarial examples can be directly attributed to the presence of non-robust features, which are highly predictive, but can be easily manipulated by adversaries to fool NLP models. In this study, we explore the feasibility of capturing task-specific robust features, while eliminating the non-robust ones by using the information bottleneck theory. Through extensive experiments, we show that the models trained with our information bottleneck-based method are able to achieve a significant improvement in robust accuracy, exceeding performances of all the previously reported defense methods while suffering almost no performance drop in clean accuracy on SST-2, AGNEWS and IMDB datasets.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑