Bibliographic Details
| Title: |
Task-Agnostic Detector for Insertion-Based Backdoor Attacks |
| Authors: |
Lyu, Weimin, Lin, Xiao, Zheng, Songzhu, Pang, Lu, Ling, Haibin, Jha, Susmit, Chen, Chao |
| Publication Year: |
2024 |
| Collection: |
Computer Science |
| Subject Terms: |
Computer Science - Computation and Language, Computer Science - Cryptography and Security |
| More Details: |
Textual backdoor attacks pose significant security threats. Current detection approaches, typically relying on intermediate feature representation or reconstructing potential triggers, are task-specific and less effective beyond sentence classification, struggling with tasks like question answering and named entity recognition. We introduce TABDet (Task-Agnostic Backdoor Detector), a pioneering task-agnostic method for backdoor detection. TABDet leverages final layer logits combined with an efficient pooling technique, enabling unified logit representation across three prominent NLP tasks. TABDet can jointly learn from diverse task-specific models, demonstrating superior detection efficacy over traditional task-specific methods.
Comment: Findings of NAACL 2024 |
| Document Type: |
Working Paper |
| Access URL: |
http://arxiv.org/abs/2403.17155 |
| Accession Number: |
edsarx.2403.17155 |
| Database: |
arXiv |
