Research map / Multi-agent systems and games / LLM agents
DualTAP: A Dual-Task Adversarial Protector for Mobile MLLM Agents
Fuyao Zhang, Jiaming Zhang, CHE WANG, Xiongtao Sun, Yurong Hao, Guowei Guan, Wenjie Li, Longtao Huang, Wei Lim
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 contrastive attention to blur perturbation (ablation) | Blur · PrivScreen | normal-task accuracy: 33% -> 54% | Improved |
| use dual-task objective with attention instead of mask+attention (ablation) | DualTAP · PrivScreen | normal-task accuracy: 52% -> over 80% | Improved |
| increase perturbation bound epsilon from 64/255 to 192/255 (ablation) | DualTAP · PrivScreen | task accuracy: 88% -> 72% | Hurt |
| increase perturbation bound epsilon from 64/255 to 192/255 (ablation) | DualTAP · PrivScreen | privacy metrics: 80~100% -> 20~40% | Improved |
| use DualTAP instead of FOA-Attack | Qwen2.5-VL-7B · PrivScreen | leakage rate: 83.81 -> 32.86 | Improved |
| use DualTAP instead of no protection | GPT-5 · PrivScreen | BERTScore: 0.9260 -> 0.1589 | Improved |
| use DualTAP instead of no protection | UI-TARS-7B · PrivScreen | task accuracy: 80.00 -> 67.00 | Hurt |
| use DualTAP instead of no protection | InternVL3-5-8B · PrivScreen | task accuracy: 83.00 -> 83.00 | No clear effect |
Closest papers by meaning
- MINIM: Privacy-Aware Minimal View for Agents via Trusted Local Sanitization (ICML 2026 · similarity 0.84)
- GUIGuard-Bench: Toward a General Evaluation for Privacy-Preserving GUI Agents (NeurIPS 2026 · similarity 0.84)
- Disrupting Hierarchical Reasoning: Adversarial Protection for Geographic Privacy in Multimodal Reasoning Models (ICLR 2026 · similarity 0.84)
- POLAR-Bench: A Diagnostic Benchmark for Privacy-Utility Trade-offs in LLM Agents (NeurIPS 2026 · similarity 0.84)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · similarity 0.84)
- CORE: Reducing UI Exposure in Mobile Agents via Collaboration Between Cloud and Local LLMs (NeurIPS 2025 · similarity 0.83)
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.
