Adversarial Incomplete Multiview Subspace Clustering Networks.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Multiview clustering enhances performance by integrating data from multiple sources.
• Incomplete multiview clustering (IMC) is a significant challenge due to missing data in real-world datasets.
• Existing IMC methods often use shallow models, ignore hidden data information, or are limited to two views.

Purpose of the Study:

• To propose a novel framework, adversarial incomplete multiview clustering (AIMC), to overcome limitations of previous IMC approaches.
• To develop a method that effectively handles missing data across multiple views.
• To improve the accuracy and robustness of multiview clustering.

Main Methods:

• The adversarial IMC (AIMC) framework is presented, focusing on learning a common latent representation for data reconstruction and missing data inference.
• Elementwise reconstruction and generative adversarial networks (GANs) are integrated for comprehensive data evaluation.
• Clustering loss is incorporated to refine the clustering structure.
• Two variants, autoencoder-based AIMC (AAIMC) and generalized AIMC (GAIMC), are explored.

Main Results:

• Experiments on six real-world datasets demonstrate the effectiveness of the proposed AIMC framework.
• Both AAIMC and GAIMC variants show superior performance compared to baseline methods.
• The framework successfully reconstructs raw data and infers missing values, capturing both overall structure and deeper semantic understanding.

Conclusions:

• The proposed AIMC framework effectively addresses the challenges of incomplete multiview clustering.
• AIMC offers a robust solution for handling missing data in multiview learning tasks.
• The developed methods, AAIMC and GAIMC, provide significant improvements over existing approaches.