Research map / Adversarial robustness and security / Backdoor and poisoning
Backdooring Rationalization
Lingxiao Kong, Jiahui Jiang, Wenchao Xu, Lei Wu
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 |
|---|---|---|---|
| use tailored BadRNL attack instead of naive BadNet/Cbat | GRU rationalization model · Fever token-level | ASR (%): 47.07 -> 100.00 (vs BadNet) | Improved |
| use tailored BadRNL attack instead of naive BadNet | GRU rationalization model · MultiRC sentence-level | ASR (%): 87.93 -> 100.00 | Improved |
| apply AttDef defense to BadRNL | GRU rationalization model · MultiRC sentence-level | ASR (%): 100.00 -> 94.37 | No clear effect |
| apply adversarial training defense to BadRNL | GRU rationalization model · Fever sentence-level | ASR (%): 100.00 -> 89.10 | Mixed |
| add BadRNL backdoor to clean model | GRU rationalization model · Fever sentence-level | clean accuracy (%): 69.65 -> 70.31 | No clear effect |
| add BadRNL backdoor to clean model | GRU rationalization model · Fever sentence-level | gold rationale F1 (%): 42.33 -> 43.06 | No clear effect |
| add rationale loss L_RAT (ablation) | BadRNL · Fever/MultiRC | gold rationale F1: up to 3.28% | Improved |
| use gradient-searched triggers instead of random triggers (ablation) | BadRNL · Fever | ASR | Improved |
| insert trigger at rationale position instead of beginning or ending (ablation) | BadRNL · Fever | ASR | Improved |
| use poison-priority sampling instead of uniform sampling (ablation) | BadRNL · five datasets | ASR | Improved |
| increase poisoning rate beyond a threshold (ablation) | BadRNL · Fever/MultiRC | ACC and GR | Hurt |
| raise poisoning rate to 0.05 (ablation) | BadRNL · Fever | ASR | Improved |
Closest papers by meaning
- Certifying Language Model Robustness with Fuzzed Randomized Smoothing: An Efficient Defense Against Backdoor Attacks (ICLR 2025 · similarity 0.88)
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models (NeurIPS 2025 · similarity 0.87)
- Defense against Backdoor Attack on Pre-trained Language Models via Head Pruning and Attention Normalization (ICML 2024 · similarity 0.87)
- BadEdit: Backdooring Large Language Models by Model Editing (ICLR 2024 · similarity 0.87)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (ICLR 2024 · similarity 0.87)
- BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models (ICLR 2024 · similarity 0.86)
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.
