Research map / Adversarial robustness and security / Black-box and transfer attacks
Towards Query-Efficient Black-Box Adversary with Zeroth-Order Natural Gradient Descent
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 ZO-NGD instead of NES-PGD | black-box adversarial attack · MNIST | average queries: 1,243 -> 523 | Helped |
| Use ZO-NGD instead of NES-PGD | black-box adversarial attack · CIFAR-10 | average queries: 417 -> 131 | Helped |
| Use ZO-NGD instead of Bandits[TD] | black-box adversarial attack · ImageNet | average queries: 791 -> 582 | Helped |
| Use ZO-NGD instead of ZOO attack | black-box adversarial attack · ImageNet | success rate: 98.9% -> 97% | Hurt |
| Use NGD instead of PGD | white-box adversarial attack · MNIST | queries: 4.88 -> 2.12 | Helped |
| Use NGD instead of PGD | white-box adversarial attack · CIFAR-10 | queries: 4.21 -> 2.06 | Helped |
| Use NGD instead of PGD | white-box adversarial attack · ImageNet | queries: 5.62 -> 2.20 | Helped |
| Set μ to different values while holding ε=0.2 (ablation) | ZO-NGD · ImageNet (Inception) | query count: 626, 582, 596 | No clear effect |
| Increase ε from 0.15 to 0.25 while holding μ=0.1 (ablation) | ZO-NGD · ImageNet (Inception) | query count: 619 -> 559 | Helped |
| Use ZO-NGD instead of NES-PGD | black-box adversarial attack with VGG16 · MNIST (VGG) | query count: 1082 -> 548 | Helped |
| Use ZO-NGD instead of NES-PGD | black-box adversarial attack with VGG16 · CIFAR-10 (VGG) | query count: 381 -> 152 | Helped |
| Use ZO-NGD instead of NES-PGD | black-box adversarial attack with VGG16 · ImageNet (VGG) | query count: 1136 -> 594 | Helped |
Closest papers by meaning
- How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective (ICLR 2022 · similarity 0.89)
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.89)
- Min-Max Optimization without Gradients: Convergence and Applications to Black-Box Evasion and Poisoning Attacks (ICML 2020 · similarity 0.88)
- A Frank-Wolfe Framework for Efficient and Effective Adversarial Attacks (AAAI 2020 · similarity 0.88)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · 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.