Research map / Adversarial robustness and security / Black-box and transfer attacks
Guessing Smart: Biased Sampling for Efficient Black-Box Adversarial Attacks
Thomas Brunner, Frederik Diehl, Michael Truong Le, Alois Knoll
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 |
|---|---|---|---|
| add Perlin low-frequency bias (ablation) | Boundary Attack · ImageNet targeted attack | success rate at 15000 queries: 0.33 -> 0.57 | Improved |
| add regional mask bias (ablation) | Boundary Attack · ImageNet targeted attack | success rate at 15000 queries: 0.33 -> 0.44 | Improved |
| add surrogate gradient bias (ablation) | Boundary Attack · ImageNet targeted attack | success rate at 15000 queries: 0.33 -> 0.38 | Improved |
| add mask bias to Perlin-biased attack (ablation) | Boundary Attack + Perlin · ImageNet targeted attack | success rate at 15000 queries: 0.57 -> 0.80 | Improved |
| add surrogate gradient bias to Perlin+mask attack (ablation) | Boundary Attack + Perlin + Mask · ImageNet targeted attack | success rate at 15000 queries: 0.80 -> 0.85 | Improved |
| add surrogate gradient bias to Perlin-biased attack (ablation) | Boundary Attack + Perlin · ImageNet targeted attack | success rate at 15000 queries: 0.57 -> 0.60 | Improved |
| add surrogate gradient bias to mask-biased attack (ablation) | Boundary Attack + Mask · ImageNet targeted attack | success rate at 15000 queries: 0.44 -> 0.49 | Improved |
| add surrogate gradient bias to Perlin+mask attack (ablation) | Boundary Attack + Perlin + Mask · ImageNet targeted attack | median queries until success: 5485 -> 5432 | Improved |
| add Perlin bias to random guessing (ablation) | Random guessing · Tiny ImageNet AVC untargeted | median L2 distance: 11.15 -> 4.28 | Improved |
| add Perlin bias to Boundary Attack (ablation) | Boundary Attack · Tiny ImageNet AVC targeted | median L2 distance: 20.2 -> 15.1 | Improved |
| add surrogate gradients to Perlin-biased attack (ablation) | Boundary Attack + Perlin · Tiny ImageNet AVC targeted | median L2 distance: 15.1 -> 9.5 | Improved |
| replace unbiased Boundary Attack with Biased Boundary Attack | Boundary Attack · ImageNet targeted attack vs other label-only attacks | success rate at 15000 queries: 0.33 -> 0.85 | Improved |
Closest papers by meaning
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.90)
- GeoDA: A Geometric Framework for Black-Box Adversarial Attacks (CVPR 2020 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.88)
- BadPart: Unified Black-box Adversarial Patch Attacks against Pixel-wise Regression Tasks (ICML 2024 · similarity 0.87)
- Amnesia as a Catalyst for Enhancing Black Box Pixel Attacks in Image Classification and Object Detection (NeurIPS 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.
