Research map / Federated learning and privacy / Differential privacy
Private Alternating Least Squares: Practical Private Matrix Completion with Tighter Rates
Steve Chien, Prateek Jain, Walid Krichene, Steffen Rendle, Shuang Song, Abhradeep Thakurta, Li Zhang
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 |
|---|---|---|---|
| use DPALS instead of DPFW | Private matrix completion on MovieLens 10M-top400 · MovieLens 10M-top400 | RMSE: 7% to 11.6% relative improvement | Improved |
| use adaptive rather than uniform sampling (ablation) | DPALS · MovieLens 10M, ε = 10 | RMSE: 0.870 -> 0.854 | Improved |
| train ALS on subsampled rather than all movie data (ablation) | Non-private ALS · MovieLens 10M, ε = 10 | RMSE: 0.785 -> 0.812 | Hurt |
| remove PSD projection (ablation) | DPALS · MovieLens 10M, ε = 10 | RMSE | Hurt |
| use adaptive rather than uniform sampling (ablation) | Non-private ALS · MovieLens 10M, ε = 10 | RMSE | No clear effect |
| replace DPFW with DPALS | Private matrix completion · Synthetic data, ε = 1 | RMSE: DPFW RMSE equal to 1; DPALS has a lower RMSE | Improved |
| replace DPFW with DPALS | Private matrix completion · Synthetic data, n = 50K | relative RMSE improvement: at least 7-fold | Improved |
| use DPALS instead of DPFW | Private matrix completion · Theoretical sample complexity for target relative error ζ | users required: DPALS: n\geq \frac{C m}{\zeta\cdot \varepsilon}; DPFW: n\geq \frac{C m^{5/4}}{\zeta^5\cdot \varepsilon} | Improved |
| use adaptive sampling to favor rare items | DPALS · MovieLens 20M | Recall@k | Improved |
Closest papers by meaning
- Differentially Private Matrix Completion Revisited (ICML 2018 · similarity 0.90)
- The Price of Privacy for Low-rank Factorization (NeurIPS 2018 · similarity 0.89)
- Sampling-Free Privacy Accounting for Matrix Mechanisms under Random Allocation (NeurIPS 2026 · similarity 0.87)
- Privacy Amplification for Matrix Mechanisms (ICLR 2024 · similarity 0.86)
- Privacy Amplification by Iteration for ADMM with (Strongly) Convex Objective Functions (AAAI 2024 · similarity 0.86)
- Near-Exact Privacy Amplification for Matrix Mechanisms (ICLR 2025 · similarity 0.86)
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.
