Research map / Adversarial robustness and security / Backdoor and poisoning
Defending Pre-trained Language Models as Few-shot Learners against Backdoor Attacks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use masking sensitivity measured by representation change relative to few-shot anchors | MDP · SST-2, MR, CR, SUBJ, TREC across BadNets, AddSent, LWP, EP, and SOS | AUC | Helped |
| Replace distributional-anchor representation change with prediction variance (ablation) | MDP · SST-2 / LWP | FAR: increases by over 50% | Hurt |
| Add the masking-invariance constraint to prompt optimization and tune its weight (ablation) | MDP · SST-2 | FAR | Mixed |
| Increase the few-shot anchor-set size (ablation) | MDP · SST-2, MR, CR, SUBJ, TREC | FAR | Helped |
| Use discrete rather than continuous prompts (ablation) | MDP · MR / BadNets | FAR: 5.60 -> 22.60 | Hurt |
| Set different training-set FRR allowances (ablation) | MDP · SST-2 | FAR | No clear effect |
Closest papers by meaning
- Does Few-Shot Learning Suffer from Backdoor Attacks? (AAAI 2024 · similarity 0.89)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.88)
- Certifying Language Model Robustness with Fuzzed Randomized Smoothing: An Efficient Defense Against Backdoor Attacks (ICLR 2025 · similarity 0.88)
- Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization (ICML 2024 · similarity 0.88)
- Backdoor Token Unlearning: Exposing and Defending Backdoors in Pretrained Language Models (AAAI 2025 · similarity 0.88)
- Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models (ICLR 2025 · similarity 0.87)
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.