Research map / Adversarial robustness and security / Black-box and transfer attacks
Exploiting the Local Parabolic Landscapes of Adversarial Losses to Accelerate Black-Box Adversarial Attack
Hoang Tran, Dan Lu, Guannan Zhang
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 |
|---|---|---|---|
| perturb in the DCT (low-frequency) domain instead of the pixel domain (ablation) | adversarial loss landscape · Inception v3, ResNet18, VGG, ResNet50 (50 images each) | parabola fit of the loss | Improved |
| use quadratic interpolation to update the iterate when queries do not improve the loss | BABIES-DCT · CIFAR-10, ImageNet, MNIST | queries | Improved |
| BABIES-DCT versus baselines on CIFAR-10 standard models | BABIES-DCT · CIFAR-10 standard models | median queries / success rate: best success rate in 3/4 tests, lowest median queries in 4/4 | Improved |
| BABIES-DCT versus Square Attack on average queries | BABIES-DCT · CIFAR-10 standard models | average queries: each leads in two tests | Mixed |
| BABIES-DCT versus baselines, targeted ImageNet | BABIES-DCT · ImageNet targeted, Inception v3 and ResNet50 | queries: 39% and 13% less than the next baseline | Improved |
| BABIES-DCT versus Square Attack, untargeted ImageNet | BABIES-DCT · ImageNet untargeted | success rate / queries | Mixed |
| BABIES-DCT versus baselines on l2-robust MNIST/CIFAR-10 | BABIES-DCT · l2-robust CIFAR-10 and MNIST | success rate: 14%-27% lead in three of four tests | Improved |
| BABIES-DCT versus baselines on l2-robust ImageNet | BABIES-DCT · l2-robust ImageNet | success rate: 18%, 29%, 19% lead | Improved |
| BABIES-DCT versus Bandits on targeted l2-robust ResNet50 | BABIES-DCT · ImageNet targeted ResNet50 (l2-robust) | queries: 50% less queries | Improved |
| BABIES-DCT against Google Cloud Vision | BABIES-DCT · Google Cloud Vision API | success rate: 82% success with 205 average queries | Improved |
Closest papers by meaning
- GeoDA: A Geometric Framework for Black-Box Adversarial Attacks (CVPR 2020 · similarity 0.91)
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.90)
- Blackbox Attacks via Surrogate Ensemble Search (NeurIPS 2022 · similarity 0.89)
- Transferable Adversarial Perturbations (ECCV 2018 · similarity 0.89)
- Guessing Smart: Biased Sampling for Efficient Black-Box Adversarial Attacks (ICCV 2019 · similarity 0.89)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.87)
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.
