Research map / Adversarial robustness and security / Black-box and transfer attacks
NATTACK: Learning the Distributions of Adversarial Examples for an Improved Black-Box Attack on Deep Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the NATTACK distribution-search approach rather than QL | Therm-adv · CIFAR10 | attack success rate (%): 42.3% -> 91.2% | Helped |
| Use the NATTACK distribution-search approach rather than QL | SAP · CIFAR10 | attack success rate (%): 96.2% -> 100% | Helped |
| Absorb projection into the objective (ablation) | QL · Therm-adv and SAP | attack success rate (%): 42.3% -> 54.7%; 96.2% -> 97.7% | Helped |
| Add transformation g (ablation) | QL with projection absorbed into objective · Therm-adv and SAP | attack success rate (%): 54.7% -> 83.3%; 97.7% -> 98.9% | Helped |
| Add z-score normalization (ablation) | QL with projection and transformation g · Therm-adv and SAP | attack success rate (%): 83.3% -> 90.9%; 98.9% -> 100% | Helped |
| Initialize NATTACK mean with the regression network (ablation) | NATTACK · ImageNet | attack success rate (%): 82% -> 91.9% | Helped |
| Initialize NATTACK mean with the regression network (ablation) | NATTACK · ImageNet | average attack runtime: 70.77s -> 48.22s | Helped |
| Use NATTACK rather than BPDA | ImageNet, Guided denoiser · ImageNet | attack success rate (%): 100.0 -> 95.5 | Hurt |
| Use NATTACK rather than BPDA | ImageNet, Randomization · ImageNet | attack success rate (%): 100.0 -> 96.5 | Hurt |
Closest papers by meaning
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.90)
- Towards Lightweight Black-Box Attack Against Deep Neural Networks (NeurIPS 2022 · similarity 0.90)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.89)
- Adversarial Example Games (NeurIPS 2020 · similarity 0.89)
- Beyond ImageNet Attack: Towards Crafting Adversarial Examples for Black-box Domains (ICLR 2022 · similarity 0.89)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.88)
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.