Consensus Graph Learning for Multi-view Clustering (IEEE TMM 2021)

Authors: Zhenglai Li, Chang Tang, Xinwang Liu, Xiao Zheng, Guanghui Yue, Wei Zhang, En Zhu

This repository contains simple Matlab and Python implementations of our paper CGL.

1. Overview

Framework of the proposed CGL method. Multi-view similarity graphs $\{\mathbf{W}^{(v)}\}_{v=1}^V$ are generated from multi-view data $\{\mathbf{X}^{(v)}\}_{v=1}^V$ in advance. Multi-view embedded representations \textcolor[rgb]{0,0,1}{$\{\mathbf{H}^{(v)}\}_{v=1}^V$} are obtained via (a) spectral embedding. To effectively capture the global consistency among multiple views, a low rank tensor $\mathcal{T}$ is learned from a corrupted tensor $\mathcal{B}$, which is constructed by stacking the inner product of normalized embedded representations $\{ \bar{\mathbf{H}}^{(v)}\bar{\mathbf{H}}^{(v)\top}\}_{v = 1}^V$ into a third-order tensor form. We further integrate the (a) spectral embedding and (b) low rank tensor representation learning into a unified optimization framework to achieve mutual promotion. Finally, the consensus graph $\mathbf{S}$ can be learned in the embedded space.

2. Usage

• Prepare the data:

  • The ORL dataset can be downloaded from Google_Drive.
  • The other datasts can be downloaded from BaiduYun(s3u3).

• Prerequisites for Python:

  • Creating a virtual environment in terminal: conda create -n CGL python=3.9
  • Installing necessary packages: pip install -r requirements.txt

• Prerequisites for Matlab:

  • Test on Matlab R2018a

• Conduct clustering

3. Citation

Please cite our paper if you find the work useful:

@article{Li_2021_CGL,
    author={Li, Zhenglai and Tang, Chang and Liu, Xinwang and Zheng, Xiao and Zhang, Wei and Zhu, En},
    journal={IEEE Transactions on Multimedia}, 
    title={Consensus Graph Learning for Multi-View Clustering}, 
    year={2022},
    volume={24},
    number={},
    pages={2461-2472},
    doi={10.1109/TMM.2021.3081930}
    }