Semi-Supervised Multiview Feature Selection With Adaptive Graph Learning.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• High-dimensional data and numerous sources necessitate effective feature selection.
• Semi-supervised multiview feature selection (SMFS) addresses learning from limited labeled and abundant unlabeled data across heterogeneous features.
• Existing methods struggle with unreliable similarity graphs and neglect projection diversity.

Purpose of the Study:

• To develop an SMFS method that simultaneously selects features and learns a unified graph structure.
• To overcome limitations of separate graph construction and feature selection in existing approaches.
• To improve robustness against noisy features and leverage diverse projection contributions.

Main Methods:

• Proposes a novel SMFS framework integrating feature projection and similarity graph learning.
• Adaptively weights and fuses feature projections to form a joint weighted projection.
• Employs implicit graph fusion to dynamically learn a compatible cross-view graph structure.
• Utilizes a convergent iterative optimization method.

Main Results:

• The proposed SMFS method effectively selects informative features and unifies graph structures.
• Demonstrates improved robustness against noisy features through dynamic graph learning.
• Achieves superior performance compared to state-of-the-art methods on various datasets.
• Preserves complementarity and consensus across original data views.

Conclusions:

• The developed SMFS method offers a significant advancement in handling multiview data.
• Simultaneous feature selection and graph learning provide a more robust and effective approach.
• The method's ability to handle noisy features and diverse projections enhances its applicability.