L0-Motivated Low-Rank Sparse Subspace (LRSSC)

Overview

MATLAB implementation of GMC-LRSSC and L0-LRSSC proposed in L0-Motivated Low-Rank Sparse Subspace Clustering. GMC-LRSSC solves subspace clustering problem by using regularization based on multivariate generalization of the minimax-concave (GMC) penalty function. L0-LRSSC solves the Schatten-0 and L0 quasi-norm regularized objective. To run proposed algorithms, example scripts and data are provided (run_dataset_name scripts).

Datasets

The datasets used in the paper can be found in the 'datasets' directory. Datasets directory includes Extended Yale B dataset from http://vision.ucsd.edu/~leekc/ExtYaleDatabase/ExtYaleB.html, the USPS dataset from https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/multiclass.html#usps, the MNIST dataset from http://yann.lecun.com/exdb/mnist/, and the ISOLET1 dataset from the UCI Machine Learning Repository (https://archive.ics.uci.edu/ml/datasets/isolet).

Citing

When using the code in your research work, please cite "ℓ₀-Motivated Low-Rank Sparse Subspace Clustering" by Maria Brbic and Ivica Kopriva.

@article{brbic2018,
title={$\ell_0$-Motivated Low-Rank Sparse Subspace Clustering},
author={Brbi\'c, Maria and Kopriva, Ivica},
journal={IEEE Transactions on Cybernetics},
year={2018},
doi={10.1109/TCYB.2018.2883566}, 
}

Acknowledgements

This work was supported by the Croatian Science Foundation (Structured Decompositions of Empirical Data for Computationally-Assisted Diagnoses of Disease) under Grant IP-2016-06-5235, and by the European Regional Development Fund (DATACROSS) under Grant KK.01.1.1.01.0009.