Shared Gaussian Process Latent Variable Model for Incomplete Multiview Clustering.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Multiview learning methods leverage complementary information from multiple data sources to enhance machine learning task performance.
• Existing methods struggle with unpaired or incomplete multiview data, especially when a significant number of instances are missing.
• Handling incomplete multiview data is challenging due to the reduced shared information available for model learning.

Purpose of the Study:

• To propose a novel shared Gaussian process (GP) latent variable model designed for incomplete multiview clustering.
• To address the limitations of existing methods in handling missing instances within multiview datasets.
• To develop a model that can effectively learn from incomplete and potentially unpaired multiview data.

Main Methods:

• A shared Gaussian process (GP) latent variable model is proposed, integrating GP advantages with multiview learning.
• The model learns aligned representative auxiliary points across individual views to compensate for missing data and enforce group-level constraints.
• Variational inference is employed for the simultaneous learning of all hyperparameters and auxiliary points.
• The formulation is naturally extendable to more than two views without increased complexity.

Main Results:

• The proposed shared GP model demonstrates superior performance in incomplete multiview clustering compared to state-of-the-art methods.
• The method effectively compensates for missing instances by learning shared information across views.
• Experimental results validate the model's effectiveness on incomplete multiview datasets.

Conclusions:

• The developed shared Gaussian process model offers a robust solution for incomplete multiview clustering.
• The approach successfully integrates the strengths of Gaussian processes and multiview learning to handle data incompleteness.
• The method provides a significant advancement in machine learning for datasets with missing information across multiple views.