Adaptive latent similarity learning for multi-view clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Existing multi-view clustering methods use original data, limiting flexibility due to noise and inter-cluster variations.
• These methods struggle when variations within clusters exceed variations between clusters.

Purpose of the Study:

• Propose a novel multi-view clustering model: adaptive latent similarity learning (ALSL).
• Address limitations of existing methods by learning a latent data representation.

Main Methods:

• ALSL utilizes an adaptively learned graph representing cluster relationships.
• Integrates latent similarity representation learning, manifold learning, and spectral clustering into a unified framework.
• Optimized efficiently using the Augmented Lagrangian Multiplier with Alternating Direction Minimization (ALM-ADM) algorithm.

Main Results:

• The latent similarity representation captures underlying cluster structures across multiple views.
• Extensive experiments on benchmark datasets demonstrate the proposed method's superiority.

Conclusions:

• ALSL offers an intuitive and efficient approach to multi-view clustering.
• The method effectively handles noise and variations in multi-view data for improved clustering performance.