Research map / Federated learning and privacy / Differential privacy
Private Frequency Estimation via Residue Number Systems
Héber Hwang Arcolezi
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 MSS decoder instead of PGR decoder | MSS · server decoding, k=22000, n=10000 | server runtime: 11.7× -> 447.8× speed-up range (Table 2) | Improved |
| use MSS instead of PGR at eps=4.5 | MSS · server decoding, eps=4.5 | server runtime (s): 56.906 -> 0.127 | Improved |
| use MSS modular encoding instead of SS | MSS · per-user message length, k=1024 and 22000 | communication bits: up to one-half in high-privacy regimes | Improved |
| use MSS instead of SS/PGR | MSS · Zipf and Spike, k=22000, n=10000 | MSE: within ≤ 1.3× | No clear effect |
| use MSS instead of GRR | MSS · Zipf and Spike, k=22000 | MSE | Improved |
| use MSS modular randomization instead of SS/GRR | MSS · Zipf, k=100 and 1024 | data reconstruction attack success | Improved |
| use MSS instead of PGR when k < projective size K | MSS · Zipf, small k | DRA | Improved |
| align evaluation domain to PGR projective size K(eps) (ablation) | PGR · Spike, base_k=100 and 1024 | DRA | Improved |
| vary number of moduli l in {3,6,9,12,15} (ablation) | MSS · Zipf, k=1024 | MSE and bit cost: l=9 can outperform l=3 and l=12 | Mixed |
| select moduli by analytical optimization (MSS[OPT]) instead of fixed l (ablation) | MSS · Zipf, k=1024 | MSE | Improved |
| compare analytical to empirical MSE (ablation) | MSS and SS · Zipf and Spike, k=1024 and 22000 | MSE | No clear effect |
| compare analytical to empirical DRA (ablation) | MSS · Zipf and Spike | DRA | Mixed |
Closest papers by meaning
- Private and Personalized Frequency Estimation in a Federated Setting (NeurIPS 2024 · similarity 0.84)
- An Efficient Private GPT Never Autoregressively Decodes (ICML 2025 · similarity 0.84)
- Improved Communication-Privacy Trade-offs in $L_2$ Mean Estimation under Streaming Differential Privacy (ICML 2024 · similarity 0.83)
- Scaling up the Banded Matrix Factorization Mechanism for Large Scale Differentially Private ML (ICLR 2025 · similarity 0.83)
- Minimax Optimal Two-Sample Testing under Local Differential Privacy (NeurIPS 2026 · similarity 0.83)
- Private Vector Mean Estimation in the Shuffle Model: Optimal Rates Require Many Messages (ICML 2024 · 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.
