Multiview Clustering via Proximity Learning in Latent Representation Space

Area of Science:

• Machine Learning
• Data Mining
• Computer Vision

Background:

• Existing multiview clustering methods often rely on the original feature space, which can be limited by redundancy and noise.
• Linear latent representation learning may fail with nonlinear data relationships, while separate nonlinear learning and clustering can lead to suboptimal representations.
• Current methods often neglect inter-cluster relations and intra-cluster correlations, impacting latent representation quality and clustering performance.

Purpose of the Study:

• To propose a novel multiview clustering method, Multiview Latent Proximity Learning (MLPL), to overcome limitations of existing approaches.
• To develop a method that simultaneously learns nonlinear latent representations and models inter- and intra-cluster data characteristics.
• To achieve improved clustering performance by integrating latent representation learning and consensus proximity learning.

Main Methods:

• MLPL learns a nonlinear latent data representation by considering both inter-cluster relations and intra-cluster correlations.
• The method simultaneously performs latent representation learning and consensus proximity learning.
• A consensus proximity matrix with k connected components is learned to directly output clustering results.

Main Results:

• MLPL effectively addresses feature redundancy and noise by learning a more robust latent representation.
• The simultaneous learning approach ensures the latent representation is well-adapted for clustering tasks.
• Experimental results on seven real-world datasets demonstrate the superior performance of MLPL compared to state-of-the-art methods.

Conclusions:

• MLPL offers a significant advancement in multiview clustering by effectively modeling latent representations and proximity.
• The method's ability to simultaneously learn representations and proximity, while considering data correlations, leads to superior clustering outcomes.
• MLPL provides a robust and effective solution for complex multiview clustering problems.