Research map / Federated learning and privacy / Differential privacy
Private-kNN: Practical Differential Privacy for Computer Vision
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Private-kNN instead of GNMAX | private semi-supervised learning · MNIST | accuracy: 98.5% -> 98.8% | Helped |
| use Private-kNN instead of GNMAX | private semi-supervised learning · SVHN | accuracy: 91.6% -> 91.6% | No clear effect |
| use Private-kNN instead of Noisy SGD | private learning · CIFAR-10 | accuracy: 70% -> 70.8% | Helped |
| use Private-kNN instead of GNMAX | multi-label attribute classification · CelebA | accuracy: 85.0% -> 85.2% | Helped |
| use Private-kNN instead of GNMAX | multi-label attribute classification · Market1501 | accuracy: 86.8% -> 89.2% | Helped |
| add a second iteration of feature refinement (ablation) | Private-kNN on SVHN · SVHN | kNN accuracy: 82.5% -> 94.41% | Helped |
| train the student CNN from first-iteration pseudo-labels (ablation) | Private-kNN on SVHN · SVHN | accuracy: 82.5% -> 86.6% | Helped |
| train the student CNN using second-iteration pseudo-labels (ablation) | Private-kNN on SVHN · SVHN | accuracy: 94.41% -> 91.6% | Hurt |
| use Poisson subsampling with Gaussian screening (ablation) | noisy screening · CIFAR-10 screening, 8192 queries | privacy cost (ε): 1.313 -> 1.04 | Helped |
| use the data-independent screening analysis instead of Gaussian mechanism screening (ablation) | PATE screening · CIFAR-10 screening, 8192 queries | privacy cost (ε): 5.67 -> 4.43 | Helped |
| use the data-dependent screening analysis instead of Gaussian mechanism screening (ablation) | PATE screening · CIFAR-10 screening, 8192 queries | privacy cost (ε): 5.67 -> 3.83 | Helped |
Closest papers by meaning
- ViP: A Differentially Private Foundation Model for Computer Vision (ICML 2024 · similarity 0.89)
- Kona: An Efficient Privacy-Preservation Framework for KNN Classification by Communication Optimization (ICML 2025 · similarity 0.87)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.86)
- Differentially private optimization for non-decomposable objective functions (ICLR 2025 · similarity 0.86)
- Private Attribute Inference from Images with Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Differentially Private Sum-Product Networks (ICML 2024 · similarity 0.85)
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.