Research map / Adversarial robustness and security / Black-box and transfer attacks
Red-teaming Retrieval-Augmented Diffusion Models via Poisoning Knowledge Bases
Xinqi Lyu, Yihao Liu, Dong Wang, Bin Xiao
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 |
|---|---|---|---|
| jointly optimize trigger for retrieval and generation (JOB) vs strongest baseline in ASR-g | JOB · RDM PLMS-based | ASR-g: 49.25 -> 54.13 | Improved |
| add retrieval reward Rrag (ablation) | JOB · RDM DDIM-based | ASR-r: 55.29 -> 80.94 | Improved |
| add generation reward Rgen (ablation) | JOB · RDM DDIM-based | ASR-g: 55.78 -> 59.78 | Improved |
| add generation reward Rgen (ablation) | JOB · RDM DDIM-based | ASR-r: 67.80 -> 80.94 | Improved |
| add fluency reward Rcoh (ablation) | JOB · RDM DDIM-based | PPL: 70.62 -> 36.69 | Improved |
| add fluency reward Rcoh (ablation) | JOB · RDM DDIM-based | ASR-g: 46.14 -> 59.78 | Improved |
| use cosine similarity instead of squared L2 for retrieval distance (ablation) | JOB · RDM DDIM-based | ASR-g: 40.80 -> 59.78 | Improved |
| use cosine similarity instead of dot product for retrieval distance (ablation) | JOB · RDM PLMS-based | ASR-r: 69.03 -> 76.54 | Improved |
| apply query rephrasing defense | JOB · RDM DDIM-based | ASR-g: 59.78 -> 49.22 | Hurt |
| apply perplexity filter defense | JOB · RDM DDIM-based | ASR-r: 80.94 -> 79.84 | No clear effect |
| JOB trigger optimization vs training-based backdoors (BadRDM) on generation success | JOB · RDM DDIM-based | ASR-g: 39.92 -> 59.78 | Improved |
Closest papers by meaning
- AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases (NeurIPS 2024 · similarity 0.87)
- RISE: Red-teaming via Iterative Strategy Evolution for Modern Text-to-Image Models (NeurIPS 2026 · similarity 0.86)
- Red-Teaming Text-to-Image Systems by Rule-based Preference Modeling (NeurIPS 2025 · similarity 0.86)
- Text-Diffusion Red-Teaming of Large Language Models: Unveiling Harmful Behaviors with Proximity Constraints (AAAI 2025 · similarity 0.86)
- RAPID: Retrieval Augmented Training of Differentially Private Diffusion Models (ICLR 2025 · similarity 0.85)
- The Stronger the Diffusion Model, the Easier the Backdoor: Data Poisoning to Induce Copyright BreachesWithout Adjusting Finetuning Pipeline (ICML 2024 · similarity 0.85)
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.
