Research map / Federated learning and privacy / Differential privacy
k-Median Clustering via Metric Embedding: Towards Better Initialization with Differential Privacy
Chenglin Fan, Ping Li, Xiaoyun Li
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 HST initialization instead of k-median++ or random initialization before local search | k-median local search (non-DP and DP) · MNIST (l1, l2) and synthetic graphs | initial k-median cost | Improved |
| follow DP-HST initialization with DP local search instead of k-median++ or random initialization | DP local search · MNIST and graphs, epsilon=1 | final k-median cost | Improved |
| use HST initialization instead of k-median++ before non-private local search | non-DP local search · MNIST | final k-median cost | No clear effect |
| demand set D imbalanced (two clusters) rather than balanced (ablation) | DP-HST · MNIST and graphs, r=1 and r=100 | initial and final cost advantage over k-median++ | Improved |
| use DP-HST initialization instead of random or k-median++ initialization | DP local search · MNIST (l1, l2) and graphs (r=1, r=100) | iterations to reach local optimum | Improved |
| use DP-HST instead of other DP algorithms | DP local search · imbalanced demand set, T=20 steps | cost of best solution found | Improved |
| use HST initialization instead of k-median++ with growing k | initialization · MNIST l2, n=2000 | running time: more efficient when k>20 | Mixed |
| increase sample size n | HST initialization vs k-median++ · MNIST l2, k=20 | running time | Hurt |
Closest papers by meaning
- Differentially Private Clustering via Maximum Coverage (AAAI 2021 · similarity 0.91)
- Online Differentially Private Consistent Clustering (NeurIPS 2026 · similarity 0.90)
- On the Price of Differential Privacy for Hierarchical Clustering (ICLR 2025 · similarity 0.89)
- Making Old Things New: A Unified Algorithm for Differentially Private Clustering (ICML 2024 · similarity 0.87)
- Average Sensitivity of Hierarchical $k$-Median Clustering (ICML 2025 · similarity 0.86)
- A Differentially Private Clustering Algorithm for Well-Clustered Graphs (ICLR 2024 · 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.
