Research map / Adversarial robustness and security / Black-box and transfer attacks
Simple Black-box Adversarial Attacks
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 confidence-guided random orthonormal basis steps instead of QL-attack | SimBA on ImageNet untargeted attacks · ImageNet untargeted | average queries: 28,174 -> 1,665 | Helped |
| Use confidence-guided random orthonormal basis steps instead of Bandits-TD | SimBA on ImageNet untargeted attacks · ImageNet untargeted | success rate: $80.5\%$ -> $98.6\%$ | Helped |
| Use confidence-guided random orthonormal basis steps instead of Bandits-TD | SimBA on ImageNet untargeted attacks · ImageNet untargeted | average L2: 5.00 -> 3.98 | Helped |
| Restrict search to low-frequency DCT directions (ablation) | SimBA-DCT on ImageNet untargeted attacks · ImageNet untargeted | average queries: 1,665 -> 1,283 | Helped |
| Restrict search to low-frequency DCT directions (ablation) | SimBA-DCT on ImageNet untargeted attacks · ImageNet untargeted | success rate: $98.6\%$ -> $97.8\%$ | Hurt |
| Restrict search to low-frequency DCT directions (ablation) | SimBA-DCT on ImageNet targeted attacks · ImageNet targeted | average L2: 9.53 -> 7.04 | Helped |
| Restrict search to low-frequency DCT directions (ablation) | SimBA-DCT on ImageNet targeted attacks · ImageNet targeted | success rate: $100\%$ -> $96.5\%$ | Hurt |
| Restrict search to low-frequency DCT directions (ablation) | SimBA-DCT on CIFAR-10 targeted attacks · CIFAR-10 targeted | average queries: 322 -> 353 | Hurt |
| Use SimBA instead of LFBA | Google Cloud Vision attack · Google Cloud Vision | success rate: 25% -> 70% | Helped |
| Use low-frequency DCT search instead of the full SimBA basis (ablation) | SimBA-DCT on ImageNet untargeted attacks · ImageNet untargeted | median queries: 944 -> 582 | Helped |
| Use low-frequency DCT search instead of the full SimBA basis (ablation) | SimBA-DCT on ImageNet targeted attacks · ImageNet targeted | median queries: 7,038 -> 4,854; approximately 2.5% failed after 60,000 queries | Mixed |
Closest papers by meaning
- Black-Box Adversarial Attack with Transferable Model-based Embedding (ICLR 2020 · similarity 0.92)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.91)
- Blackbox Attacks via Surrogate Ensemble Search (NeurIPS 2022 · similarity 0.91)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.89)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.89)
- DifAttack: Query-Efficient Black-Box Adversarial Attack via Disentangled Feature Space (AAAI 2024 · similarity 0.89)
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.