Lujia Shen (Zhejiang University), Yuwen Pu (Zhejiang University), Shouling Ji (Zhejiang University), Changjiang Li (Penn State), Xuhong Zhang (Zhejiang University), Chunpeng Ge (Shandong University), Ting Wang (Penn State)

Transformer-based models, such as BERT and GPT, have been widely adopted in natural language processing (NLP) due to their exceptional performance. However, recent studies show their vulnerability to textual adversarial attacks where the model's output can be misled by intentionally manipulating the text inputs. Despite various methods that have been proposed to enhance the model's robustness and mitigate this vulnerability, many require heavy consumption resources (e.g., adversarial training) or only provide limited protection (e.g., defensive dropout).

In this paper, we propose a novel method called dynamic attention, tailored for the transformer architecture, to enhance the inherent robustness of the model itself against various adversarial attacks. Our method requires no downstream task knowledge and does not incur additional costs. The proposed dynamic attention consists of two modules: *(i) attention rectification*, which masks or weakens the attention value of the chosen tokens, and *(ii) dynamic modeling*, which dynamically builds the set of candidate tokens. Extensive experiments demonstrate that dynamic attention significantly mitigates the impact of adversarial attacks, improving up to 33% compared to previous methods against widely used adversarial attacks. The model-level design of dynamic attention enables it to be easily combined with other defense methods (e.g., adversarial training) to further enhance the model's robustness. Furthermore, we demonstrate that dynamic attention preserves the state-of-the-art robustness space of the original model compared to other dynamic modeling methods.

View More Papers

CamPro: Camera-based Anti-Facial Recognition

Wenjun Zhu (Zhejiang University), Yuan Sun (Zhejiang University), Jiani Liu (Zhejiang University), Yushi Cheng (Zhejiang University), Xiaoyu Ji (Zhejiang University), Wenyuan Xu (Zhejiang University)

Read More

Experimental Analyses of the Physical Surveillance Risks in Client-Side...

Ashish Hooda (University of Wisconsin-Madison), Andrey Labunets (UC San Diego), Tadayoshi Kohno (University of Washington), Earlence Fernandes (UC San Diego)

Read More

Overconfidence is a Dangerous Thing: Mitigating Membership Inference Attacks...

Zitao Chen (University of British Columbia), Karthik Pattabiraman (University of British Columbia)

Read More

Detecting Voice Cloning Attacks via Timbre Watermarking

Chang Liu (University of Science and Technology of China), Jie Zhang (Nanyang Technological University), Tianwei Zhang (Nanyang Technological University), Xi Yang (University of Science and Technology of China), Weiming Zhang (University of Science and Technology of China), NengHai Yu (University of Science and Technology of China)

Read More