Research map / Adversarial robustness and security / Black-box and transfer attacks
Generalizable Data-Free Objective for Crafting Universal Adversarial Perturbations
Konda Reddy Mopuri, Aditya Ganeshan, R. Venkatesh Babu
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 range prior (mean and dynamic range) (ablation) | GD-UAP on GoogLeNet · ILSVRC fooling rate | fooling rate (%): 58.62 -> 71.44 | Improved |
| add training data samples as prior (ablation) | GD-UAP on GoogLeNet · ILSVRC fooling rate | fooling rate (%): 71.44 -> 83.54 | Improved |
| add training data samples as prior (ablation) | GD-UAP on ResNet-152 · ILSVRC fooling rate | fooling rate (%): 37.3 -> 66.68 | Improved |
| use data w/ less background instead of all data (ablation) | GD-UAP on FCN-8s-VGG · Pascal VOC segmentation | GFR (%): 61.15 -> 67.19 | Improved |
| use data w/ less background instead of all data (ablation) | GD-UAP on DL-RN101 · Pascal VOC segmentation | GFR (%): 58.62 -> 56.61 | Hurt |
| use l2-norm activation objective instead of mean activation (ablation) | GD-UAP with data on CaffeNet · ILSVRC fooling rate | fooling rate (%): 88.35 -> 91.54 | Improved |
| use UAP crafted on mismatched data (Places-205 to ILSVRC) | UAP on GoogLeNet · ILSVRC fooling rate | fooling rate (%): 78.5 -> 28.17 | Hurt |
| use range prior noise instead of real data for UAP | UAP on GoogLeNet · ILSVRC fooling rate | fooling rate (%): 78.5 -> 10.56 | Hurt |
| add data prior over range prior on depth estimation (ablation) | GD-UAP on Monodepth-ResNet50 · KITTI GFR on delta<1.25 | GFR (%): 47.6% -> 24.3% | Hurt |
| apply PRN defense against UAP perturbations | GD-UAP data prior on GoogLeNet · ILSVRC fooling rate | fooling rate (%): 83.54 -> 52.50 | Hurt |
| use scattering-based hybrid network instead of ResNet-18 | GD-UAP transfer attack from VGG-13, range prior · ILSVRC fooling rate | fooling rate (%): 39.14 -> 30.96 | Hurt |
Closest papers by meaning
- Defense Against Universal Adversarial Perturbations (CVPR 2018 · similarity 0.91)
- Data-free Universal Adversarial Perturbation with Pseudo-semantic Prior (CVPR 2025 · similarity 0.90)
- Data-Free Universal Attack by Exploiting the Intrinsic Vulnerability of Deep Models (AAAI 2025 · similarity 0.89)
- Generate Universal Adversarial Perturbations for Few-Shot Learning (NeurIPS 2024 · similarity 0.89)
- Improving Generalization of Universal Adversarial Perturbation via Dynamic Maximin Optimization (AAAI 2025 · similarity 0.88)
- Democratic Training Against Universal Adversarial Perturbations (ICLR 2025 · 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.
