Timezone: »

A Unified Evaluation of Textual Backdoor Learning: Frameworks and Benchmarks
Ganqu Cui · Lifan Yuan · Bingxiang He · Yangyi Chen · Zhiyuan Liu · Maosong Sun

Thu Dec 08 09:00 AM -- 11:00 AM (PST) @

Textual backdoor attacks are a kind of practical threat to NLP systems. By injecting a backdoor in the training phase, the adversary could control model predictions via predefined triggers. As various attack and defense models have been proposed, it is of great significance to perform rigorous evaluations. However, we highlight two issues in previous backdoor learning evaluations: (1) The differences between real-world scenarios (e.g. releasing poisoned datasets or models) are neglected, and we argue that each scenario has its own constraints and concerns, thus requires specific evaluation protocols; (2) The evaluation metrics only consider whether the attacks could flip the models' predictions on poisoned samples and retain performances on benign samples, but ignore that poisoned samples should also be stealthy and semantic-preserving. To address these issues, we categorize existing works into three practical scenarios in which attackers release datasets, pre-trained models, and fine-tuned models respectively, then discuss their unique evaluation methodologies. On metrics, to completely evaluate poisoned samples, we use grammar error increase and perplexity difference for stealthiness, along with text similarity for validity. After formalizing the frameworks, we develop an open-source toolkit OpenBackdoor to foster the implementations and evaluations of textual backdoor learning. With this toolkit, we perform extensive experiments to benchmark attack and defense models under the suggested paradigm. To facilitate the underexplored defenses against poisoned datasets, we further propose CUBE, a simple yet strong clustering-based defense baseline. We hope that our frameworks and benchmarks could serve as the cornerstones for future model development and evaluations.

Author Information

Ganqu Cui (Tsinghua University, Tsinghua University)
Lifan Yuan (Huazhong University of Science and Technology)

I am a final-year undergraduate student in School of Artificial Intelligence and Automation, Huazhong University of Science and Technology. My research interests lie in trustworthy NLP systems, improving their security and robustness and establishing appropriate evaluation frameworks for them.

Bingxiang He (清华大学)
Yangyi Chen (Huazhong University of Science and Technology)
Zhiyuan Liu (Tsinghua University)
Maosong Sun (Tsinghua University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors