https://github.com/ourownstory/federated_kmeans

Federated k-means clustering algorithm implementation and proof of concept.
https://github.com/ourownstory/federated_kmeans

clustering federated-learning kmeans kmeans-clustering machine-learning privacy

Last synced: 29 days ago
JSON representation

Federated k-means clustering algorithm implementation and proof of concept.

• Host: GitHub
• URL: https://github.com/ourownstory/federated_kmeans
• Owner: ourownstory
• License: mit
• Created: 2019-04-26T16:41:09.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2021-09-07T09:39:05.000Z (about 3 years ago)
• Last Synced: 2024-05-12T00:45:23.336Z (6 months ago)
• Topics: clustering, federated-learning, kmeans, kmeans-clustering, machine-learning, privacy
• Language: Python
• Homepage:
• Size: 531 MB
• Stars: 25
• Watchers: 2
• Forks: 2
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Federated K-Means Clustering Algorithm

You can find the details on the algorithm and a proof of concept in the [short paper](federated_kmeans_arxiv.pdf)

## Abstract

An algorithm to cluster distributed datasets without communicating data is introduced. It builds upon the concept of federated learning, distributed K-Means and mini-batch K-Means. Results on synthetic data and real data are presented for a non-iid setting. The algorithm is able to produce a clustering of similar or better quality than central K-Means clustering, while preserving privacy.

## Results

The Federated clustering algorithm is able to match or outperform the central clustering algorithm if the hyperparameters are appropriately set. 

Further preliminary experiments on real energy datasets indicated that the Algorithm is simple and effective to apply to industrial problem settings (results not included). 

## Note

This is unpolished work, done in May to September 2019. Nevertheless, we decided to share the findings already with the broader science community as others might find use in our simple algorithm.