arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2204.05239cs.CL
收藏

探究自主学习范式的普遍脆弱性

Exploring the Universal Vulnerability of Prompt-based Learning Paradigm

Tsinghua University(清华大学);Beijing National Research Center for Information Science and Technology(北京国家信息科学与技术研究中心);Huazhong University of Science and Technology(华中科技大学);Chongqing University(重庆大学)

Lei Xu, Yangyi Chen, Ganqu Cui, Hongcheng Gao, Zhiyuan Liu

摘要:

基于自适应的学习范式弥补了预训练和微调之间的差距,并在Few-Shot设置下有效地工作。然而,我们发现这种学习范式继承了预训练阶段的脆弱性,在预训练阶段,模型预测可能会被插入某些触发器到文本中所误导。在本文中,我们通过注入后门触发器或仅使用纯文本在预训练的语言模型上搜索对抗触发器来探索这种普遍的漏洞。在这两种情况下,我们证明,我们的触发器可以完全控制或严重降低性能微调任意下游任务,反映了普遍的脆弱性,基于神经网络的学习范式。进一步的实验表明,对抗触发器在语言模型之间具有良好的可移植性。我们还发现,传统的微调模型不容易受到从预先训练的语言模型构建的对抗性触发器的影响。最后,我们提出了一个潜在的解决方案,以减轻我们的攻击方法。代码和数据可在https://github.com/leix28/prompt-universal-vulnerability上公开获取

英文摘要:

Prompt-based learning paradigm bridges the gap between pre-training and fine-tuning, and works effectively under the few-shot setting. However, we find that this learning paradigm inherits the vulnerability from the pre-training stage, where model predictions can be misled by inserting certain triggers into the text. In this paper, we explore this universal vulnerability by either injecting backdoor triggers or searching for adversarial triggers on pre-trained language models using only plain text. In both scenarios, we demonstrate that our triggers can totally control or severely decrease the performance of prompt-based models fine-tuned on arbitrary downstream tasks, reflecting the universal vulnerability of the prompt-based learning paradigm. Further experiments show that adversarial triggers have good transferability among language models. We also find conventional fine-tuning models are not vulnerable to adversarial triggers constructed from pre-trained language models. We conclude by proposing a potential solution to mitigate our attack methods. Code and data are publicly available at https://github.com/leix28/prompt-universal-vulnerability

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。

↑