Research map / Adversarial robustness and security / Backdoor and poisoning
A Unified Evaluation of Textual Backdoor Learning: Frameworks and Benchmarks
Ganqu Cui, Lifan Yuan, Bingxiang He, Yangyi Chen, Zhiyuan Liu, Maosong Sun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply CUBE filtering | BERT-base poisoned training · HSOL BadNet | ASR: 99.84 -> 100.0 | No clear effect |
| apply ONION/BKI/STRIP/RAP token-level defenses | BERT-base · SST-2 SynBkd and StyleBkd | ASR: 86.08 -> 88.49 (BKI SynBkd) | No clear effect |
| apply inference-time RAP defense | BERT-base · SST-2/HSOL/AG's News | CACC: 91.10 -> 45.74 (SST-2 clean) | Hurt |
| apply ONION inference-time defense | BERT-base · SST-2 BadNet | ASR: 100.0 -> 20.50 | Mixed |
| increase poison rate | dataset-releasing attacks · SST-2/HSOL/AG's News | ASR: near 100% ASR at poison rate >=0.1 | Improved |
| use clean-label instead of mix/dirty-label poisoning | dataset-releasing attacks · SST-2/HSOL/AG's News | ASR | Hurt |
| fine-tune on larger clean dataset IMDB after poisoning on SST-2 | RIPPLES · SST-2 -> IMDB | ASR: 100 -> 16.81 | Hurt |
| use three triggers instead of one | POR · HSOL pre-trained model | ASR: 11.21 -> 68.49 | Improved |
| use three triggers instead of one | NeuBA · SST-2 pre-trained model | ASR: 73.40 -> 65.25 | Hurt |
Closest papers by meaning
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.92)
- Backdoor Token Unlearning: Exposing and Defending Backdoors in Pretrained Language Models (AAAI 2025 · similarity 0.91)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.90)
- Certifying Language Model Robustness with Fuzzed Randomized Smoothing: An Efficient Defense Against Backdoor Attacks (ICLR 2025 · similarity 0.90)
- Simulate and Eliminate: Revoke Backdoors for Generative Large Language Models (AAAI 2025 · similarity 0.90)
- Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization (ICML 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
