Research map / Adversarial robustness and security / Certified and empirical robustness
Mind the Box: $l_1$-APGD for Sparse Adversarial Attacks on Image Classifiers
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 |
|---|---|---|---|
| exact projection onto l1-ball ∩ [0,1]^d instead of approximate projection (L1+clipping) (ablation) | l1-APGD and existing l1-PGD attacks (SLIDE, Maini et al.) · CIFAR-10 l1 eps=12, Figure 2 | robust accuracy and best loss | Helped |
| adaptive sparsity and step-size schedule (single-eps l1-APGD) instead of fixed top-k sparsity (SLIDE) | l1-PGD attack · CIFAR-10 l1 eps=12, 100 iterations, Table 1 | robust accuracy (%): APGD_CE 61.3 vs SLIDE 66.6 on the APGD-AT model; 50.4 vs 53.8 on Maini AVG | Helped |
| multi-eps schedule (3eps, 2eps, eps phases) added to l1-APGD (ablation) | single-eps l1-APGD · CIFAR-10 l1 eps=12, Figure 2 | robust loss and accuracy | Helped |
| upscale Square candidate by 3 and project onto l1-ball ∩ [0,1]^d (l1-Square adaptation) | Square Attack · CIFAR-10 l1 eps=12 black-box comparison, Table 4 | robust accuracy (%) | Helped |
| adversarial training with l1-APGD (10 steps) instead of standard l1-PGD | ResNet-18 adversarial training, l1 eps=12 · CIFAR-10 high budget, Table 2 | robust accuracy (%) under AA: 60.3 vs 51.9 for best prior model (Madaan et al.) | Helped |
| reduce APGD steps from 100 to 25 (single-eps, adaptive) (ablation) | single-eps l1-APGD · CIFAR-10 l1 eps=12, Figure 3 | robust accuracy | No clear effect |
Closest papers by meaning
- Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations (ICML 2024 · similarity 0.90)
- Towards Stable and Efficient Adversarial Training against $l_1$ Bounded Adversarial Attacks (ICML 2023 · similarity 0.88)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.88)
- Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack (ICML 2020 · similarity 0.88)
- Stop Walking in Circles! Bailing Out Early in Projected Gradient Descent (CVPR 2025 · similarity 0.87)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.86)
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.