Welcome to the world of deletion efficient AI systems.

This repository contains a research prototype of two provably deletion efficient k-means algorithms, proposed by tginart et al. in a forthcoming publication at NeurIPS 2019.

If you use this prototype for research, please reference the original paper Making AI forget you: Data deletion in machine learning:

@inproceedings{ginart2019making,
  title={Making AI forget you: Data deletion in machine learning},
  author={Ginart, Antonio and Guan, Melody and Valiant, Gregory and Zou, James Y},
  booktitle={Advances in Neural Information Processing Systems},
  pages={3513--3526},
  year={2019}
}

Please see demo.ipynb for a tutorial on using the code. The source code can be found in del_eff_kmeans.py. Once you've gone through the demo, you can try out our algorithms on real datasets! See datasets for the preprocessed data used in the paper. All of the data is publicly available as well (see paper for references).