Research map / Adversarial robustness and security / Black-box and transfer attacks
RaPA: Enhancing Transferable Targeted Attacks via Random Parameter Pruning
Tongrui Su, Qingbin Li, Shengyu Zhu, Wei Chen, Xueqi Cheng
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 |
|---|---|---|---|
| apply DropConnect only to BN and FC layers rather than all layers (ablation) | RaPA · 16 targets, RN50 surrogate | Average ASR (%): 69.2 -> 72.4 | Improved |
| apply DropConnect to Conv layers instead of BN layers (ablation) | RaPA · 16 targets, RN50 surrogate | Average ASR (%): 72.1 -> 65.1 | Hurt |
| apply DropConnect only to the FC layer in ResNet-50 (ablation) | RaPA · 16 targets, RN50 surrogate | Average ASR (%): 72.4 -> 41.1 | Hurt |
| apply DropConnect to LN plus FC layers instead of one type for ViT (ablation) | RaPA · 16 targets, ViT surrogate | Average ASR (%): 58.6 -> 65.2 | Improved |
| vary DropConnect probability p from 0.01 to 0.09 (ablation) | RaPA · 16 targets, RN50 surrogate | Average ASR (%): mean 66.3, peak 72.4 at p = 0.05 | Mixed |
| increase iterations 300 to 500 and inferences per iteration 1 to 5 (ablation) | RaPA · 16 targets, RN50 surrogate | Average ASR (%): +15.9 | Improved |
| add RaPA to a DSM-trained surrogate | DSM · Transformer targets, RN50 surrogate | Average ASR (%): 20.6 -> 58.3 | Improved |
| add RaPA to a SASD-WS surrogate (no weight scaling) | SASD-WS · Transformer targets, RN50 surrogate | Average ASR (%): 42.2 -> 51.6 | Improved |
| use RaPA instead of the best baseline on Transformer targets with RN50 surrogate | targeted transfer attack · Transformer targets, RN50 surrogate | Average ASR (%): 33.3 -> 45.0 | Improved |
| use RaPA instead of the best baseline on Transformer targets with DN121 surrogate | targeted transfer attack · Transformer targets, DN121 surrogate | Average ASR (%): 22.8 -> 40.3 | Improved |
| use RaPA against strong defenses ensIR | targeted transfer attack · Defense ensIR, RN50 surrogate | ASR (%): 13.8 -> 43.2 | Improved |
| use RaPA against Diffpure | targeted transfer attack · Defense Diffpure, RN50 surrogate | ASR (%): 4.0 | No clear effect |
Closest papers by meaning
- Boosting the Transferability of Adversarial Attacks with Reverse Adversarial Perturbation (NeurIPS 2022 · similarity 0.90)
- Seeking Flat Minima Over Diverse Surrogates for Improved Adversarial Transferability: A Theoretical Framework and Algorithmic Instantiation (TPAMI 2026 · similarity 0.89)
- Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024 · similarity 0.88)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.87)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.87)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 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.
