Research map / Federated learning and privacy / Differential privacy
Differentially Private Cross-Silo Recommendation from Implicit Feedback
Xun Ran, Qingqing Ye, Xin Huang, Jianliang Xu, Haibo Hu
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 complementary loss and symmetric noise to strawman objective perturbation (ablation) | DPIMF · ML-10M | HR@10: DPIMFstr exhibits the worst performance due to high-variance noise | Improved |
| add symmetric noise (upper-triangular perturbation) to the complementary loss (ablation) | DPIMFcom · ML-10M (alpha0=0.8) | HR: 2.2% higher average HR | Improved |
| set alpha0=1 (no second-order perturbation) (ablation) | DPIMFsym · ML-10M | HR@10: performs best across all eps | Improved |
| add Poisson importance sampling (privacy amplification) to DPIMF (ablation) | DPIMF · ML-10M, HR@10 vs eps | HR@10: advantage more pronounced for eps < 5 | Improved |
| use privacy-aware importance sampling instead of uniform sampling (ablation) | DPIMF-IS · ML-10M (2,000 users, d=20) | sampling weights / variance of linear coefficient: weights closely match utility-optimal solution; uniform markedly different | Improved |
| replace DPLCF with DPIMF-IS | private item profile on ML-10M · ML-10M | HR: 18.3% | Improved |
| replace LDPICF with DPIMF-IS | private item profile on ML-10M · ML-10M | HR: 8.3% | Improved |
| replace DPLCF with DPIMF-IS | private item profile on ML-10M · ML-10M | NDCG: 12.1% | Improved |
| replace LDPICF with DPIMF-IS | private item profile on ML-10M · ML-10M | NDCG: 5.4% | Improved |
| replace SGDMF with DPIMF-IS (one-shot objective perturbation vs DP-SGD) | DP recommender · ML-10M, eps=0.5 | HR@10: 0.15 -> 0.33 | Improved |
| replace EFVAE+DP-Adam with DPIMF-IS | DP recommender · ML-10M, eps=0.5 | HR@10: 0.11 -> 0.33 | Improved |
| replace SGDMF with DPIMF-IS | DP recommender · ML-10M, eps=10 | HR@10: 0.48 -> 0.54 | Improved |
Closest papers by meaning
- On Privacy and Personalization in Cross-Silo Federated Learning (NeurIPS 2022 · similarity 0.86)
- Differentially Private Matrix Completion Revisited (ICML 2018 · similarity 0.85)
- Differentially Private Sparse Reward Estimation with Preference Feedback (NeurIPS 2026 · similarity 0.84)
- RRL: Recommendation Reverse Learning (AAAI 2024 · similarity 0.84)
- Reducing Item Discrepancy via Differentially Private Robust Embedding Alignment for Privacy-Preserving Cross Domain Recommendation (ICML 2024 · similarity 0.84)
- Unbiased Recommender Learning from Implicit Feedback via Weakly Supervised Learning (ICML 2025 · 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.
