Awesome

EMC-Nets

The official repos of Efficient Multi-view Clustering Networks

Abstract

Deep learning has made remarkable progress on multi-view clustering (MvC) in the last decade. Most existing literature adopted a broad target to guide network learning, such as minimize the reconstruction loss. It is effective but not efficient. In this paper, we proposed a novel framework, named Efficient Multi-view Clustering Networks (EMC-Nets), which guarantees the efficiency of the network learning and produces discriminative common representation efficiently in multiple sources. Specifically, the proposed method alternates between the instruction process and approximation process during training. The instruction process employs a standard clustering algorithm, such as k-means, to generate pseudo-labels corresponding to the current common representation. The approximation process leverages pseudo-labels to force the network to approximate a reasonable cluster distribution. Experimental results on four real-world datasets demonstrate that the proposed method outperforms state-of-the-art methods.

Architecture

architecture

Attention Fusion Layer

attn

Environment setting

python 3.7
pytorch 1.8.1
CUDA 10.2

We recommend using Conda to setup the environment, and run as the following:

Create the virtual environment and install the requirements.

conda create -n EMC-Nets python=3.7
conda activate EMC-Nets
cd EMC-Nets
conda install --yes --file requirements.txt

Then, use unittest to test this project, following:

cd tests
export PYTHONPATH="../"
python -m unittest

Quickly validation

Run:

python validation.py

Result:

------------------------------ Begin validation BDGP ------------------------------
CV 1 Beginning...
Acc.: 98.60% NMI: 95.41% purity: 98.60% and ARI: 96.55%
CV 2 Beginning...
Acc.: 98.56% NMI: 95.27% purity: 98.56% and ARI: 96.45%
CV 3 Beginning...
Acc.: 98.56% NMI: 95.27% purity: 98.56% and ARI: 96.45%
CV 4 Beginning...
Acc.: 98.56% NMI: 95.27% purity: 98.56% and ARI: 96.45%
CV 5 Beginning...
Acc.: 98.56% NMI: 95.27% purity: 98.56% and ARI: 96.45%
After ron 5 times, final Acc.: 98.57% NMI: 95.30% purity: 98.57% and ARI: 96.47%
------------------------------ End validation BDGP ------------------------------

More training details see logs/

Training

Coming soon.

Visualization

Here, we present the visualization of BDGP. More details see our paper, please.

visualization

Citation

@article{Ke_2022,	
doi = {10.1007/s10489-021-03129-0},	
url = {https://doi.org/10.1007/s10489-021-03129-0},	
year = 2022,	
month = {jan},	
publisher = {Springer Science and Business Media {LLC}},	
author = {Guanzhou Ke and Zhiyong Hong and Wenhua Yu and Xin Zhang and Zeyi Liu},	
title = {Efficient multi-view clustering networks},	
journal = {Applied Intelligence}}

Acknowledge

This paper was inspired by DeepCluster, OnlineCluster.