Research map / Federated learning and privacy / Privacy attacks and secure learning
Boundary Probing for Input Privacy Protection When Using LMM Services
Xiaofei Hui, Haoxuan Qu, Ping Hu, Hossein Rahmani, Jun Liu
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 GEP probing (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | action accuracy: 24.1 -> 47.9 | Improved |
| add GEP probing (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | privacy cMAP: 57.3 -> 54.3 | Improved |
| use PGP instead of stopping enhancement at the boundary (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | privacy cMAP: 61.0 -> 54.3 | Improved |
| use PGP instead of reducing the learning rate (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | privacy cMAP: 60.7 -> 54.3 | Improved |
| add PGP versus stop enhancing (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | action accuracy: 47.6 -> 47.9 | No clear effect |
| add PGP versus reducing learning rate (ablation) | PABP · UCF101-VISPR (Video-LLaVA) | action accuracy: 47.5 -> 47.9 | No clear effect |
| use PABP instead of SelectivePrivacy | Video-LLaVA · UCF101-VISPR (Video-LLaVA) | action accuracy: 39.8 -> 47.9 | Improved |
| use PABP instead of SelectivePrivacy | Video-LLaVA · UCF101-VISPR (Video-LLaVA) | privacy cMAP: 59.1 -> 54.3 | Improved |
| use PABP instead of SelectivePrivacy | GPT-4V · HMDB51-VISPR (GPT-4V) | action accuracy: 49.5 -> 55.1 | Improved |
| use PABP instead of SelectivePrivacy | LLaVA · OK-VQA (LLaVA) | VQA accuracy: 47.1 -> 57.3 | Improved |
| use PABP instead of SelectivePrivacy | GPT-4V · OK-VQA (GPT-4V) | privacy cMAP: 51.5 -> 43.6 | Improved |
| anonymize with PABP versus raw data | Video-LLaVA · HMDB51-VISPR (Video-LLaVA) | action accuracy: 50.3 -> 46.7 | Hurt |
Closest papers by meaning
- PID: Prompt-Independent Data Protection Against Latent Diffusion Models (ICML 2024 · similarity 0.84)
- Validating Privacy-Preserving Face Recognition under a Minimum Assumption (CVPR 2024 · similarity 0.84)
- ObCLIP: Oblivious CLoud-Device Hybrid Image Generation with Privacy Preservation (NeurIPS 2025 · similarity 0.84)
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language Models (ICML 2026 · similarity 0.82)
- Portcullis: A Scalable and Verifiable Privacy Gateway for Third-Party LLM Inference (AAAI 2025 · similarity 0.82)
- 𝑓-Differential Privacy Filters: Validity and Approximate Solutions (NeurIPS 2026 · similarity 0.82)
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.
