Research map / Adversarial robustness and security / Attacks on text and graphs
AdvDiff: Generating Unrestricted Adversarial Examples using Diffusion Models
Xuelong Dai, Kaisheng Liang, 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 |
|---|---|---|---|
| use diffusion-based AdvDiff instead of GAN-based U-GAN on MNIST LeNet5 against PGD-AT | unrestricted adversarial attack · MNIST LeNet5 (PGD-AT) | ASR (%): 79.4 -> 88.6 | Improved |
| replace PGD-gradient guidance with AdvDiff adversarial guidance in diffusion attacks | diffusion adversarial sampling · ImageNet generation quality | FID: 20.5 -> 16.2 (DiffAttack vs AdvDiff) | Improved |
| AdvDiff instead of U-BigGAN on generation quality | unrestricted adversarial attack · ImageNet generation quality | IS: 129.4 -> 343.8 | Improved |
| AdvDiff instead of DiffAttack/AdvDiffuser on generation time | diffusion attacks · ImageNet ResNet50 | Time (s): 28.2 -> 9.2 (vs DiffAttack) | Improved |
| use targeted AdvDiff for black-box transfer vs earlier attacks | AdvDiff · ImageNet transfer to ResNet-152 | ASR (%): 32.5 -> 20.5 (AutoAttack vs AdvDiff) | Hurt |
| use untargeted AdvDiff for transfer and defenses | AdvDiff · ImageNet against DiffPure | ASR (%): 41.6 -> 75.2 | Improved |
| increase noise sampling guidance steps N (ablation) | AdvDiff · ImageNet ResNet50 | ASR vs generation quality | Mixed |
| set adversarial guidance scale extremely large (ablation) | AdvDiff · ImageNet ResNet50 | image realism | Hurt |
| apply adversarial guidance in early reverse timesteps (ablation) | AdvDiff · ImageNet ResNet50 | ASR | No clear effect |
Later papers that compared with this method
2 later accepted papers report AdvDiff as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- ObjectAdv: Object-Level Unrestricted Adversarial Attacks via Diffusion Models (AAAI 2026 · better on 4 of 4 shared results)
- V-Attack: Targeting Disentangled Value Features for Controllable Adversarial Attacks on LVLMs (CVPR 2026 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DiffAttack — DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification (NeurIPS 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AdvDiffuser: Natural Adversarial Example Synthesis with Diffusion Models (ICCV 2023 · similarity 0.92)
- DiffAdvMAP: Flexible Diffusion-Based Framework for Generating Natural Unrestricted Adversarial Examples (ICML 2025 · similarity 0.89)
- A Diffusion Prior-based Framework for Imperceptible and Flexible Unrestricted Adversarial Attacks (TPAMI 2026 · similarity 0.89)
- ADBM: Adversarial Diffusion Bridge Model for Reliable Adversarial Purification (ICLR 2025 · similarity 0.88)
- AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (NeurIPS 2024 · similarity 0.88)
- Adv-Diffusion: Imperceptible Adversarial Face Identity Attack via Latent Diffusion Model (AAAI 2024 · 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.
