Research map / Adversarial robustness and security / Black-box and transfer attacks
Query-Efficient Black-Box Adversarial Attacks Guided by a Transfer-Based Prior
Yinpeng Dong, Shuyu Cheng, Tianyu Pang, Hang Su, Jun Jie Zhu
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 transfer-based prior via biased sampling | RGF · ImageNet l2 (Inception-v3) | average queries: 1309 -> 745 | Improved |
| add transfer-based prior via gradient averaging | RGF · ImageNet l2 (Inception-v3) | average queries: 1309 -> 735 | Improved |
| use optimal lambda* instead of fixed lambda=0.05 (ablation) | PRGF-BS · ImageNet l2 (Inception-v3) | average queries: 1047 -> 745 | Improved |
| use optimal mu* instead of fixed mu=0.5 (ablation) | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 958 -> 735 | Improved |
| use optimal mu* instead of fixed mu=0.5 (ablation) | PRGF-GA · CIFAR-10 l2 (ResNet-50) | average queries: 306 -> 173 | Improved |
| add data-dependent prior | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 735 -> 644 | Improved |
| add data-dependent prior | RGF · ImageNet l2 (Inception-v3) | average queries: 1309 -> 910 | Improved |
| add data-dependent prior with three surrogate models | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 191 -> 267 | Hurt |
| use two surrogates with equal averaging instead of one (ablation) | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 735 -> 575 | Improved |
| use subspace projection instead of equal averaging for two surrogates (ablation) | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 575 -> 342 | Improved |
| add third surrogate Inception-ResNet-v2 (ablation) | PRGF-GA · ImageNet l2 (Inception-v3) | average queries: 342 -> 191 | Improved |
| use Inception-v4 instead of ResNet-v2-152 surrogate (ablation) | PRGF-GA · ImageNet l2 (ResNet-50) | average queries: 250 -> 350 | Hurt |
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.
- NATTACK — NATTACK: Learning the Distributions of Adversarial Examples for an Improved Black-Box Attack on Deep Neural Networks (ICML 2019) · better on 3 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
- Improving Black-box Adversarial Attacks with a Transfer-based Prior (NeurIPS 2019 · similarity 0.98)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.93)
- Practical Black-box Attacks on Deep Neural Networks using Efficient Query Mechanisms (ECCV 2018 · similarity 0.92)
- Boosting Black-Box Attack With Partially Transferred Conditional Adversarial Distribution (CVPR 2022 · similarity 0.92)
- Prior Convictions: Black-box Adversarial Attacks with Bandits and Priors (ICLR 2019 · similarity 0.92)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.91)
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.
