Research map / Adversarial robustness and security / Black-box and transfer attacks
Diversity can be Transferred: Output Diversification for White- and Black-box Attacks
Yusuke Tashiro, Yang Song, Stefano Ermon
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 |
|---|---|---|---|
| Initialize PGD with 2-step ODI instead of naive uniform restarts | PGD · CIFAR-10 | accuracy: 46.06\pm 0.02\% -> 44.45\pm 0.02\% | Improved |
| Initialize C&W with ODI instead of Gaussian initialization | C&W · ImageNet | average minimum ℓ2 perturbation: 1.58\pm0.00 -> 1.32\pm0.01 | Improved |
| Use tuned ODI-PGD rather than MultiTargeted | CIFAR-10 robust model · CIFAR-10 | accuracy: 44.03\% -> 44.00 \pm 0.01\% | Improved |
| Replace SimBA-DCT directions with ODS directions from surrogate models | SimBA · ImageNet untargeted | average queries: 909 -> 242 | Improved |
| Replace SimBA-DCT directions with ODS directions from surrogate models | SimBA · ImageNet targeted | median ℓ2 perturbation: 7.00 -> 3.55 | Improved |
| Replace RGF random sampling with ODS using one surrogate | RGF · ImageNet ℓ2 untargeted | average queries: 633 -> 133 | Improved |
| Replace RGF random sampling with ODS using one surrogate | RGF · ImageNet ℓ∞ targeted | success rate: 65.3\% -> 92.0\% | Improved |
| Use ODS-RGF instead of Square Attack | Black-box score-based attacks · ImageNet ℓ2 untargeted | average queries: 647 -> 144 | Improved |
| Use ODS-RGF instead of Square Attack | Black-box score-based attacks · ImageNet ℓ∞ untargeted | average queries: 60 -> 78 | Hurt |
| Replace Boundary Attack Gaussian directions with ODS from four surrogates | Boundary Attack · ImageNet untargeted, 10000 queries | median ℓ2 perturbation: 4.30 -> 0.57 | Improved |
| Replace Boundary Attack Gaussian directions with ODS from four surrogates | Boundary Attack · ImageNet targeted, 10000 queries | median ℓ2 perturbation: 27.05 -> 3.76 | Improved |
| Train ODS surrogate models on an OOD dataset rather than the full dataset (ablation) | Boundary-ODS · ImageNet untargeted, 10000 queries | median ℓ2 perturbation: 0.57 -> 0.98 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Square Attack — Square Attack: a query-efficient black-box adversarial attack via random search (ECCV 2020) · better on 1 of 2 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 Transferability of Adversarial Examples With Input Diversity (CVPR 2019 · similarity 0.88)
- Guessing Smart: Biased Sampling for Efficient Black-Box Adversarial Attacks (ICCV 2019 · similarity 0.87)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.87)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.86)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.85)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 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.
