Research map / Adversarial robustness and security / Backdoor and poisoning
Unleashing Stealthy Backdoor Pandemic by Infecting a Single Diffusion Model
Mohaiminul Al Nahian, Abeer Matar Almalky, Sabbir Ahmed, Abdullah Al Arafat, Mamshad Nayeem Rizve, Adnan Siraj Rakin
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 |
|---|---|---|---|
| add synthetic images from the attacked SD model to limited real labels | CIFAR-10 classifiers · CIFAR-10 (8% labels) | accuracy: 80.01 -> 82.85 (ResNet-20) | Improved |
| add synthetic images from the attacked SD model to limited real labels | CIFAR-100 classifiers · CIFAR-100 (8% labels) | accuracy: 40.33 -> 49.23 (ResNet-20) | Improved |
| add synthetic images from the attacked SD model to limited real labels | TinyImageNet classifiers · TinyImageNet (8% labels) | accuracy: 34.54 -> 43.55 (ResNet-18) | Improved |
| put the pre-trained VAE in the trigger optimization loop (ablation) | Eidolon trigger optimization · Trigger fidelity after UNet training | trigger distribution match | Improved |
| use Eidolon instead of prior diffusion backdoors (SBA, BadT2I, TPA) | Stable Diffusion data augmentation · CIFAR-10, ResNet-20 | ASR (%): 12.89 -> 99.66 (best prior BadT2I Pixel vs Eidolon) | Improved |
| swap the UNet-based SD for MMDiT SD-3-medium (ablation) | Eidolon · CIFAR-10, ResNet-20 | ASR (%): 99.66 -> 91.30 | Hurt |
| optimize trigger against a zero-shot classifier | Eidolon triggered images · BLIP label check | target label predicted rate: 97.22% | Improved |
| use ViT/Swin instead of CNNs as downstream classifier | Eidolon on TinyImageNet · TinyImageNet (8% labels) | ASR (%): 97.78 (ResNet-18) -> 90.67 (ViT-B), 88.44 (Swin-T) | Hurt |
Closest papers by meaning
- Elijah: Eliminating Backdoors Injected in Diffusion Models via Distribution Shift (AAAI 2024 · similarity 0.89)
- TERD: A Unified Framework for Safeguarding Diffusion Models Against Backdoors (ICML 2024 · similarity 0.88)
- BackdoorDM: A Comprehensive Benchmark for Backdoor Learning on Diffusion Model (NeurIPS 2025 · similarity 0.87)
- UFID: A Unified Framework for Black-box Input-level Backdoor Detection on Diffusion Models (AAAI 2025 · similarity 0.87)
- UIBDiffusion: Universal Imperceptible Backdoor Attack for Diffusion Models (CVPR 2025 · similarity 0.87)
- Demystifying Poisoning Backdoor Attacks from a Statistical Perspective (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.
