This is the implementation of our paper t-k-means: A Robust and Stable k-means Variant, accepted by the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021. The project is developed based on the MATLAB, created by Yang Zhang and Yiming Li. In this paper, we propose a novel robust and stable k-means variant, the t-k-means, and its fast version based on the understanding of k-means, GMM, and TMM.

Citation

If our work is useful for your research, please cite our paper as follows:

@inproceedings{li2021t,
  title={t-k-means: A Robust and Stable k-means Variant}
  author={Li, Yiming and Zhang, Yang and Tang, Qingtao and Huang, Weipeng and Jiang, Yong and Xia, Shu-Tao},
  booktitle={ICASSP},
  year={2021}
}

Description of Main Codes

• main.m: test algorithms on a specified dataset and generate results.
• gmmCluster.m: the implement of GMM algorithm.
• tmmCluster.m: the implement of TMM algorithm.
• kmeansCluster.m: the implement of k-means algorithm.
• kmeansppCluster.m: the implement of k-means++ algorithm.
• kmedianCluster.m: the implement of k-median algorithm.
• kmedoidCluster.m: the implement of k-medoid algorithm.
• SigmaAlphaCluster.m: the implement of t-k-means algorithm.
• Sigma0Cluster.m: the implement of fast t-k-means algorithm.
• Sigma0ppCluster.m: the implement of fast t-k-means++ algorithm.
• plot_loss.m: loss visualization.

Evaluation

run

main.m

with different parameters (i.e., data_version, method_count, and exp_count) and method names to generate needed results.