Research map / Adversarial robustness and security / Black-box and transfer attacks
PGA: Prior-free Generative Attack for Practical No-box Scenario
Hongyu Peng, Xiang Yuan, Gong 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 |
|---|---|---|---|
| add Curriculum-Guided Micro-Robust Optimization to surrogate training (ablation) | BIA generator · ImageNet cross-model (CNNs) | AVGc ASR (%): 54.18 -> 65.44 | Improved |
| add Region-Aware Consistent Perturbation Learning to generator training (ablation) | BIA generator · ImageNet cross-model (CNNs) | AVGc ASR (%): 54.18 -> 62.11 | Improved |
| combine CMRO and RCPL (ablation) | PGA · ImageNet cross-model (ViTs/MLPs) | AVGv ASR (%): 32.95 -> 47.50 | Improved |
| increase number of regions K from 1 to 4 (ablation) | RCPL · ImageNet cross-model (CNNs) | AVGc ASR (%): 65.44 -> 71.39 | Improved |
| increase number of regions K from 4 to 9 (ablation) | RCPL · ImageNet cross-model (ViTs/MLPs) | AVGv ASR (%): 47.50 -> 42.29 | Hurt |
| delay micro-robust training until epoch 200 with tau_max 0.01 instead of other strength or timing (ablation) | CMRO · ImageNet cross-model | AVGc and AVGv ASR | Mixed |
| use generative attacks with SimSiam surrogate instead of iterative PNS attacks | PNS attacks · ImageNet cross-model | ASR: AGS AVGc 50.26 vs BIA 54.18 | Improved |
| add DA or RN variants to BIA | BIA · ImageNet cross-model (CNNs) | AVGc ASR (%): 54.18 -> 54.32 (DA), 54.51 (RN) | No clear effect |
| use PGA instead of strongest baseline at eps=10 | generative attack · ImageNet cross-model (eps=10) | AVGc ASR (%): 34.22 -> 50.11 (BIA vs PGA) | Improved |
| use PGA instead of baselines against defenses | generative attack · ImageNet under defenses | AVGd ASR (%): 56.88 -> 72.16 (FACL vs PGA) | Improved |
| reduce perturbation budget from 16 to 15 without retraining (ablation) | PGA · ImageNet image quality | AVGc ASR (%): 71.39 -> 68.95 | Hurt |
| use PGA instead of baselines for cross-task attack | Faster R-CNN · MS COCO detection (ImageNet-trained generator) | AP (%): 21.35 -> 17.60 (GAPF vs PGA) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CDTA — CDTA: A Cross-Domain Transfer-Based Attack with Contrastive Learning (AAAI 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- On Generating Transferable Targeted Perturbations (ICCV 2021 · similarity 0.88)
- Improving Black-Box Generative Attacks via Generator Semantic Consistency (ICLR 2026 · similarity 0.88)
- AIM: Additional Image Guided Generation of Transferable Adversarial Attacks (AAAI 2025 · similarity 0.87)
- Attack Deterministic Conditional Image Generative Models for Diverse and Controllable Generation (AAAI 2024 · similarity 0.87)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · similarity 0.86)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 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.
