Research map / Adversarial robustness and security / Backdoor and poisoning
Exposing Functional Fusion: A New Class of Strategic Backdoor in Dynamic Prompt Architectures
Zeyao Liu, Zhendong Zhao, Xiaojun Chen, Xin Zhao, Yuexin Xuan, Xiaoshuang Ji
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 |
|---|---|---|---|
| use dynamic context-aware prompt generator instead of static VPT prompts (ablation) | VIPER · Average over 6 datasets | ACC (%): 82.81 -> 87.77 | Improved |
| use dynamic prompts instead of static prompts (ablation) | VIPER · Food101 | ACC (%): 76.56 -> 89.95 | Improved |
| keep the Core (0.27% of parameters) in addition to Periphery (ablation) | VIPER · ImageNet100 | ASR (%): 1.52 -> 100.00 | Improved |
| keep only the Core, dropping the Periphery (ablation) | VIPER · ImageNet100 | ACC (%): 91.44 -> 91.10 | No clear effect |
| perturbative fine-tuning with poisoned images and random labels for one epoch (ablation) | VIPER · ImageNet100 | ACC (%): 91.44 -> 2.52 | Hurt |
| add VIPER VPG to frozen ViT-B/16 | Clean ViT · ImageNet100 | ACC (%): 86.16 -> 91.44 | Improved |
| increase prompt length from 8 to 16 (ablation) | VIPER · ImageNet100 and DTD | ACC (%): ImageNet100 91.44 -> 90.50; DTD 75.23 -> 76.85 | Mixed |
| decrease prompt length from 8 to 4 (ablation) | VIPER · UCF101 | ACC (%): 82.37 -> 80.61 | Hurt |
| inject VPG at all 12 layers instead of layers 3, 6, 9 (ablation) | VIPER · ImageNet100 | ACC (%): 91.44 -> 91.92 | No clear effect |
| inject VPG at shallow and middle layers 1-8 instead of 3, 6, 9 (ablation) | VIPER · ImageNet100 | ACC (%): 91.44 -> 92.44 | No clear effect |
| replace LoRA attack with VIPER under 90% pruning | VIPER vs LoRA attack · ImageNet100 pruning defense | ASR (%): 100 at 90% pruning; LoRA collapses after 60% | Improved |
| train on ImageNet100 and transfer zero-shot to five unseen datasets | VIPER · Cross-dataset transfer | ACC (%): 83.13 -> 81.43 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TrojViT — TrojViT: Trojan Insertion in Vision Transformers (CVPR 2023) · better on 2 of 2 shared results
- WaNet — WaNet - Imperceptible Warping-based Backdoor Attack (ICLR 2021) · better on 2 of 2 shared results
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
- BadVLA: Towards Backdoor Attacks on Vision-Language-Action Models via Objective-Decoupled Optimization (NeurIPS 2025 · similarity 0.87)
- Revisiting Backdoor Attacks against Large Vision-Language Models from Domain Shift (CVPR 2025 · similarity 0.86)
- Mitigating the Backdoor Effect for Multi-Task Model Merging via Safety-Aware Subspace (ICLR 2025 · similarity 0.85)
- Everyday Object Meets Vision-and-Language Navigation Agent via Backdoor (NeurIPS 2024 · similarity 0.85)
- Attention-Imperceptible Backdoor Attacks on Vision Transformers (AAAI 2025 · similarity 0.85)
- Not All Prompts Are Secure: A Switchable Backdoor Attack Against Pre-trained Vision Transfomers (CVPR 2024 · similarity 0.84)
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.
