Research map / Adversarial robustness and security / Black-box and transfer attacks
Blackbox Attacks via Surrogate Ensemble Search
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 |
|---|---|---|---|
| increase surrogate ensemble size from 4 to 20 (ablation) | BASES · ImageNet targeted attack | fooling rate | Helped |
| add coordinate-descent search over ensemble weights (vs equal weights) (ablation) | BASES N=20 · VGG-19 targeted | fooling rate: 54% -> 96% | Helped |
| use black-box query coordinate descent instead of exact white-box gradient for weight update (ablation) | BASES · ImageNet targeted attack | fooling rate | No clear effect |
| use weighted-loss ensemble instead of weighted probabilities or logits (ablation) | BASES PM · VGG-19 targeted | fooling rate | No clear effect |
| use cross-entropy instead of C&W margin loss (ablation) | BASES PM · VGG-19 targeted | fooling rate | No clear effect |
| shuffle order of surrogate models (ablation) | BASES · VGG-19 targeted | fooling rate | No clear effect |
| vary learning rate eta from 0.001 to 0.02 (ablation) | BASES · VGG-19 targeted | fooling rate | No clear effect |
| increase PGD step size to 3x the I-FGSM step (ablation) | BASES PM · VGG-19 targeted | fooling rate | Helped |
| replace TREMBA/GFCS with BASES | targeted blackbox attack · ImageNet VGG-19 | fooling rate with fewer queries: 89.2% -> 95.9% | Helped |
| choose hardest target label instead of NeurIPS17 label | BASES · DenseNet-121 targeted | fooling rate: 100% -> 97% | Hurt |
Closest papers by meaning
- Learning Black-Box Attackers with Transferable Priors and Query Feedback (NeurIPS 2020 · similarity 0.91)
- Attacking deep networks with surrogate-based adversarial black-box methods is easy (ICLR 2022 · similarity 0.91)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.91)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.89)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.89)
- Amnesia as a Catalyst for Enhancing Black Box Pixel Attacks in Image Classification and Object Detection (NeurIPS 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.