A survey on representation learning for multi-view data.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view clustering leverages information from diverse data sources.
• Existing surveys often overlook the integration of self-supervised and non-self-supervised approaches.
• The field has seen significant growth in recent decades.

Purpose of the Study:

• To provide a novel survey and classification of multi-view clustering algorithms.
• To categorize existing methods into self-supervised and non-self-supervised frameworks.
• To offer an insightful overview of developments in multi-view clustering.

Main Methods:

• Classification into non-self-supervised and self-supervised multi-view clustering.
• Review of non-self-supervised methods: non-representation learning (matrix factorization, kernel) and representation learning (graph, deep, subspace).
• Review of self-supervised methods: contrastive and generative approaches.

Main Results:

• A structured categorization of multi-view clustering techniques.
• Detailed examination of diverse algorithms within each category.
• Identification of key methodologies in both supervised and unsupervised learning paradigms.

Conclusions:

• The survey consolidates existing knowledge in multi-view clustering.
• It highlights the importance of considering both self-supervised and non-self-supervised methods.
• Provides a valuable resource for researchers in machine learning and data mining.