Exploring Clean Label Backdoor Attacks and Defense in Language Models
- 5 June 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE/ACM Transactions on Audio, Speech, and Language Processing
- Vol. 32 (23299290) , 3014-3024
Abstract
Despite being widely applied, pre-trained language models have been proven vulnerable to backdoor attacks. Backdoor attacks are designed to introduce targeted vulnerabilities into models by poisoning a subset of training samples through trigger injection and label modification. Traditional textual backdoor attacks suffer several flaws: the triggers lead to abnormal natural language expressions, and poisoned sample labels are mistakenly labeled. These flaws reduce the stealthiness of the attack and can be easily detected by defense models. In this study, we introduce Cbat, a novel and efficient method to perform clean-label backdoor attack with text style, which does not require external trigger, and the poisoned samples are correctly labeled. Specifically, we develop a sentence rewriting model by leveraging the powerful few-shot learning capability of prompt tuning to generate clean label poisoned samples. Cbat then injects text style as an abstract trigger into the victim model through poisoned samples. We also introduce an algorithm for defending against backdoor attacks, named CbatD, which effectively erases the poisoned samples by locating the lowest training loss and calculating feature relevance. The experiments on text classification tasks demonstrate that our Cbat and CbatD show overall competitive performance in textual backdoor attack and defense. It is noteworthy that Cbat attained leading results in the clean-label backdoor attack benchmark without triggers.Keywords
Funding Information
- National Natural Science Foundation of China (12271215, 12326378, 11871248)
- China Scholarship Council (202206780011)
- Outstanding Innovative Talents Cultivation Funded Programs for Doctoral Students of Jinan University (2022CXB013)
This publication has 67 references indexed in Scilit:
- Not All Samples Are Born Equal: Towards Effective Clean-Label Backdoor AttacksPattern Recognition, 2023
- Towards Backdoor Attacks and Defense in Robust Machine Learning ModelsComputers & Security, 2023
- Kallima: A Clean-Label Framework for Textual Backdoor Attacks2022
- 2D Target/Anomaly Detection in Time Series Drone Images Using Deep Few-Shot Learning in Small Training Dataset2022
- Mitigating backdoor attacks in LSTM-based text classification systems by Backdoor Keyword IdentificationNeurocomputing, 2021
- A Prior-mask-guided Few-shot Learning for Skin Lesion SegmentationComputing, 2021
- Few-shot learning creates predictive models of drug response that translate from high-throughput screens to individual patientsNature Cancer, 2021
- Few-Shot Learning for Medical Image Classification2020
- Clean-Label Backdoor Attacks on Video Recognition Models2020
- An implementation of optimized framework for action classification using multilayers neural network on selected fused featuresPattern Analysis and Applications, 2018