Differentially Private Federated k-Means Clustering with Server-Side Data

This repository provides an implementation of the ICML paper, "Differentially Private Federated k-Means Clustering with Server-Side Data".

Installation

Clone the repository and install the required dependencies:

pip install -r requirements.txt

Usage

Run the experiments via the run.py script. Below are the usage instructions for different privacy settings and datasets.

---

Data Point-Level Privacy

Mixture of Gaussians Dataset

The following command runs FedDP-KMeans on a mixture of Gaussians with 100 total clients in a data-point-level privacy setting.

python run.py --args_config configs/gaussians_data_privacy.yaml

Folktables Dataset

The following command runs FedDP-KMeans on the folktables dataset in a data-point-level privacy setting.

python run.py --args_config configs/folktables.yaml

---

Client-Level Privacy

Mixture of Gaussians Dataset

The following command runs FedDP-KMeans on a mixture of Gaussians with 2000 total clients in a client-level privacy setting.

python run.py --args_config configs/gaussians_client_privacy.yaml

StackOverflow Dataset

For StackOverflow first download and process the dataset using the following commands (takes some time to run). Set NJPS to parallelize, in total 3*NJPS threads will be run.

cd ./data/stackoverflow
. process_stackoverflow.sh $NJPS

After preprocessing the dataset, use the following command to run FedDP-KMeans on the dataset with topic tags github and pdf, for a total of 9237 clients.

python run.py --args_config configs/stackoverflow.yaml

---

Citation

If you use this code or refer to our work, please cite the following paper:

@inproceedings{scott2025clustering,
  author    = {Scott, Jonathan and Lampert, Christoph H and Saulpic, David},
  title     = {Differentially Private Federated $k$-Means Clustering with Server-Side Data},
  booktitle = {Forty-second International Conference on Machine Learning, {ICML}},
  year      = {2025},
  url       = {https://arxiv.org/abs/2506.05408},
}