Multi-view Clustering

A categorization and reproduction for Multi-view Clustering(MvC).

Contents

• Papers
• Performance

Papers

MvC for consensus learning

These methods adopt different strategies to collect information effectively from each view. Indies of clustering performance (i.e. Accuracy, NMI) is mainly used to measure the quality of the model. In general, they are mainly based on matrix factorization, spectral-graph theory and others like kernel mapping and tensor rank constraint. Note that differences of these methods are reflected in the assumption on data distribution. Matrix factorization are applied to solve linear seperatable data, while spectral-graph theory and kernel mapping methods are good at handling non-linear ones.

Graph-based method

• CAN Clustering and Projected Clustering with Adaptive Neighbors [code]
  • MLAN Multi-View Clustering and Semi-Supervised Classification with Adaptive Neighbours [code]
• CLR The constrained laplacian rank algorithm for graph-based clustering [code]
  • SwMC Self-weighted Multiview Clustering with Multiple Graphs [code]
  • PwMC Self-weighted Multiview Clustering with Multiple Graphs [code]
• Co-reg Co-regularized multi-view spectral clustering [code]
• MCGC Multiview Consensus Graph Clustering [code]

Spectral-based method

• AMGL Parameter-free auto-weighted multiple graph learning: a framework for multiview clustering and semi-supervised classification [code]
• AWP Multiview Clustering via Adaptively Weighted Procrustes
• Uni A Unified Weight Learning Paradigm for Multi-view Learning

MvC for large-scale data

• MVSC Large-scale multi-view spectral clustering via bipartite graph
• SMKMC Multi-View K-Means Clustering on Big Data

Incomplete MvC

Partial consensus Learning

• PVC Partial multi-view clustering
• SRLCSimultaneous Representation Learning and Clustering for Incomplete Multi-view Data code
• IMCRVIncomplete Multi-View Clustering with Reconstructed Views [unreleased code]

Data completion

• PIC Spectral Perturbation Meets Incomplete Multi-view Data[code]
• AGC-IMC Adaptive Graph Completion Based Incomplete Multi-view Clustering[code]

Performance

MvC for consensus learning

		Mfeat			Caltech101-7
Methods	Acc	NMI	F-measure	Acc	NMI	F-measure
SC_best	0.829	0.843	0.802	0.636	0.518	0.282
CLR_best	0.832	0.838	0.802	0.712	0.526	0.267
SwMC	0.858	0.888	0.838	0.651	0.582	0.319
PwMC	0.860	0.885	0.839	0.652	0.548	0.320
CAN_best	0.876	0.904	0.844	0.723	0.435	0.248
MLAN	0.952	0.913	0.954	0.816	0.550	0.379
Co-reg	0.964	0.931	0.964	0.651	0.550	0.308
AMGL	0.876	0.894	0.861	0.674	0.650	0.392