Research map / Adversarial robustness and security / Black-box and transfer attacks
DASH: A Meta-Attack Framework for Synthesizing Effective and Stealthy Adversarial Examples
Abdullah Al Nomaan Nafi, Habibur Rahaman, Zafaryab Haider, Tanzim Mahfuz, Fnu Suya, Swarup Bhunia, Prabuddha Chakraborty
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 |
|---|---|---|---|
| train attack-combination weights instead of using random weights (ablation) | DASH · CIFAR-100, Cui2024 | ASR (no post-processing, %): 80.00 -> 100.00 | Improved |
| train attack-combination weights instead of using random weights (ablation) | DASH · CIFAR-100, Cui2024 | SSIM (x100): 90.86 -> 94.43 | Improved |
| chain two stages instead of one (ablation) | DASH · CIFAR-100, Cui2024 | average ASR (%): 51.88 -> 99.49 | Improved |
| chain three stages instead of two (ablation) | DASH · CIFAR-100, Cui2024 | average ASR (%): 99.49 -> 99.77 | No clear effect |
| increase stages from 3 to 5 (ablation) | DASH · CIFAR-100, Cui2024 | SSIM (x100): 94.43 -> 92.92 | Hurt |
| increase stages from 3 to 5 (ablation) | DASH · CIFAR-100, Cui2024 | average ASR (%): 99.77 -> 99.98 | Improved |
| use only one stage versus more stages (ablation) | DASH · CIFAR-100, Cui2024 | SSIM (x100): 98.48 (1 stage) vs 94.43 (3 stages) | Mixed |
| compose multiple base attacks with learned weights versus the best single attack | DASH vs CW · CIFAR-100, Cui2024 | average ASR (%): 84.14 -> 99.77 | Improved |
| use DASH instead of AdvAD on ImageNet robust model | Salman2020 ImageNet-1k · ImageNet-1k, Salman2020 | SSIM (x100): 71.30 -> 96.52 | Improved |
| transfer learned weights across models and datasets (ablation) | DASH · cross-model weight transfer | average ASR (%) | No clear effect |
| use DASH surrogate on Addepalli2022 instead of AdvAD for transfer | DASH · CIFAR-100 transfer, surrogate Addepalli2022 | average transfer ASR (%): 64.70 -> 63.30 | No clear effect |
| use DASH instead of DiffAttack for transfer from Cui2024 | DASH · CIFAR-100 transfer, surrogate Cui2024 | average transfer ASR (%): 74.74 -> 84.62 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AdvAD — AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (NeurIPS 2024) · better on 8 of 9 shared results
- DiffAttack — DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification (NeurIPS 2023) · better on 6 of 6 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
- Chain of Attack: On the Robustness of Vision-Language Models Against Transfer-Based Adversarial Attacks (CVPR 2025 · similarity 0.86)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.86)
- Enhancing Diffusion-based Unrestricted Adversarial Attacks via Adversary Preferences Alignment (NeurIPS 2025 · similarity 0.86)
- Exploring Visual Vulnerabilities via Multi-Loss Adversarial Search for Jailbreaking Vision-Language Models (CVPR 2025 · similarity 0.85)
- Steering Away from Harm: An Adaptive Approach to Defending Vision Language Model Against Jailbreaks (CVPR 2025 · similarity 0.85)
- Towards Understanding and Improving Adversarial Robustness of Vision Transformers (CVPR 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.
