Research map / Federated learning and privacy / Differential privacy
Privacy-Preserving Visual Learning Using Doubly Permuted Homomorphic Encryption
Ryo Yonetani, Vishnu Naresh Boddeti, Kris M. Kitani, Yoichi Sato
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 |
|---|---|---|---|
| increase sparsity of local classifier updates from 0.01% to 95.6% (ablation) | DPHE · Caltech101 | accuracy (%): 89.7 -> 88.2 | No clear effect |
| increase sparsity of local classifier updates from 0.01% to 95.6% (ablation) | DPHE · Caltech101 | encryption time (sec): 620 -> 62 | Improved |
| increase sparsity from 0.01% to 79.1% (ablation) | DPHE · Caltech101 | encryption time (sec): 620 -> 186 | Improved |
| increase number of users N from 5 to 100 with fixed, partially duplicated images per user (ablation) | DPHE · Caltech101 | accuracy (%) | No clear effect |
| increase number of users N from 5 to 100 with non-duplicated images split among users (ablation) | DPHE · Caltech101 | accuracy (%) | Hurt |
| use DPHE (no added noise, iterative averaging) instead of RA12 | privacy-preserving classifier learning · Caltech101 | accuracy (%): 83.8 -> 89.3 | Improved |
| use DPHE instead of PRR10 | privacy-preserving classifier learning · Caltech101 | accuracy (%): 41.6 -> 89.3 | Improved |
| use DPHE instead of RA12 | privacy-preserving classifier learning · Caltech256 | accuracy (%): 68.0 -> 74.7 | Improved |
| use DPHE instead of RA12 | privacy-preserving classifier learning · CelebA (40 attributes) | mean accuracy (%): 64 -> 84 | Improved |
| use DPHE instead of PRR10 | privacy-preserving classifier learning · CelebA (40 attributes) | mean accuracy (%): 78 -> 84 | Improved |
| use DPHE instead of RA12 | privacy-preserving classifier learning · Life-logging sensitive place detection | average precision: 0.704 -> 0.729 | Improved |
| add privacy preservation via DPHE relative to non-private No-PP SVM | linear SVM · YouTube8M subset (top 100 attributes) | mAP: 0.48 -> 0.47 | No clear effect |
Closest papers by meaning
- ENSEI: Efficient Secure Inference via Frequency-Domain Homomorphic Convolution for Privacy-Preserving Visual Recognition (CVPR 2020 · similarity 0.87)
- Revisiting ML Training under Fully Homomorphic Encryption: Convergence Guarantees, Differential Privacy, and Efficient Algorithms (ICML 2026 · similarity 0.86)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.85)
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · similarity 0.83)
- PPIDSG: A Privacy-Preserving Image Distribution Sharing Scheme with GAN in Federated Learning (AAAI 2024 · similarity 0.83)
- Privacy-Preserving Embedding via Look-up Table Evaluation with Fully Homomorphic Encryption (ICML 2024 · 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.
