Research map / Adversarial robustness and security / Attacks on text and graphs
AdvDrop: Adversarial Attack to DNNs by Dropping Information
Ranjie Duan, Yuefeng Chen, Dantong Niu, Yun Yang, A. K. Qin, Yuan He
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 |
|---|---|---|---|
| relax constraint epsilon from 20 to 100 (ablation) | AdvDrop · ImageNet targeted | success rate: 97.20 -> 99.95 | Improved |
| relax constraint epsilon from 20 to 100 (ablation) | AdvDrop · ImageNet untargeted | success rate: 98.55 -> 100.00 | Improved |
| relax epsilon from 20 to 100 (ablation) | AdvDrop targeted · ImageNet | steps to reach 99% success: 496 -> 61 | Improved |
| use proposed differentiable quantization instead of typical rounding (ablation) | AdvDrop · ImageNet | success rate: 5.00 -> 97.20 | Improved |
| use proposed differentiable quantization instead of Shin et al. rounding (ablation) | AdvDrop · ImageNet | success rate: 65.20 -> 97.20 | Improved |
| drop information in frequency domain instead of reducing color depth to 4 bits (ablation) | AdvDrop · ImageNet | accuracy: 93.50 -> 11.10 | Improved |
| drop low frequency region (ablation) | clean images · ImageNet | accuracy: 100.00 -> 30.00 | Hurt |
| drop middle frequency region (ablation) | clean images · ImageNet | accuracy: 100.00 -> 84.40 | Hurt |
| drop high frequency region (ablation) | clean images · ImageNet | accuracy: 100.00 -> 86.50 | Hurt |
| use AdvDrop instead of PGD l_inf under JPEG-30 defense | PGD · ImageNet | attack success rate: 62.50 -> 80.00 | Improved |
| use AdvDrop instead of PGD l_inf under pixel deflection | PGD · ImageNet | attack success rate: 85.10 -> 95.65 | Improved |
| use AdvDrop instead of PGD l_inf under adversarial training | PGD · ImageNet | attack success rate: 41.60 -> 44.50 | No clear effect |
Closest papers by meaning
- AdvDiffuser: Natural Adversarial Example Synthesis with Diffusion Models (ICCV 2023 · similarity 0.87)
- A Direct Approach to Robust Deep Learning Using Adversarial Networks (ICLR 2019 · similarity 0.87)
- Deep Defense: Training DNNs with Improved Adversarial Robustness (NeurIPS 2018 · similarity 0.87)
- Adversarial Attacks Are Reversible With Natural Supervision (ICCV 2021 · similarity 0.87)
- AdvAD: Exploring Non-Parametric Diffusion for Imperceptible Adversarial Attacks (NeurIPS 2024 · similarity 0.86)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.85)
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.
