Research map / Adversarial robustness and security / Black-box and transfer attacks
Black-Box Adversarial Attack with Transferable Model-based Embedding
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 |
|---|---|---|---|
| search in the pretrained adversarial generator's embedding space instead of the input space | TREMBA · MNIST, untargeted | queries: 4380 -> 1064 | Helped |
| search in the pretrained adversarial generator's embedding space instead of the input space | TREMBA · ImageNet, targeted class 0 (tench), VGG19 | queries: 12283 -> 853 | Helped |
| train the embedding generator on adversarial patterns from a source model rather than use an unsupervised reconstruction autoencoder | TREMBA · ImageNet, targeted class 0 (tench), VGG19 | success rate: 25.80% -> 98.47% | Helped |
| optimize the latent starting point on the defended source network (ablation) | TREMBA · ImageNet defended ResNeXt101 DenoiseAll | success rate: 49.59% -> 50.41% | Helped |
| optimize the latent starting point on the defended source network (ablation) | TREMBA · CIFAR10 defended ResNeXt | queries: 2528 -> 4994 | Hurt |
| make AutoZOOM's embedding space smaller (ablation) | AutoZOOM · ImageNet, untargeted | success rate: 64.35% (VGG19); 71.94% (Resnet34); 68.44% (DenseNet121); 71.15% (MobilenetV2) | Hurt |
| vary NES sample size b (ablation) | TREMBA · ImageNet DenseNet121, untargeted | queries: 150 (b=10); 205 (b=30); 274 (b=40); 299 (b=50) | Mixed |
| use additional source networks in the ensemble (ablation) | TREMBA · ImageNet targeted attack | success rate | Helped |
| use the generator trained at ε=0.03125 at other perturbation bounds (ablation) | TREMBA · ImageNet targeted attack, ε=0.02 and ε=0.04 | success rate | Helped |
Closest papers by meaning
- Learning Black-Box Attackers with Transferable Priors and Query Feedback (NeurIPS 2020 · similarity 0.92)
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.92)
- 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 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.89)
- Ask, Attend, Attack: An Effective Decision-Based Black-Box Targeted Attack for Image-to-Text Models (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.