Fast multi-view clustering via correntropy-based orthogonal concept factorization.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Concept factorization (CF) enhances multi-view clustering by handling negative data.
• Existing CF methods are sensitive to feature dimensions and noise, impacting effectiveness.
• Standard CF can suffer from non-unique factorization, reducing clustering performance.

Purpose of the Study:

• To develop a fast and robust multi-view clustering method addressing limitations of existing CF approaches.
• To improve efficiency by reducing sensitivity to high feature dimensions.
• To enhance robustness against complex noise and ensure unique factorization.

Main Methods:

• Factorization is performed on a learned consensus anchor graph, not the original data space.
• A lightweight graph regularization term refines factorization with low computational cost.
• An orthogonal CF model incorporating the correntropy criterion is developed for improved robustness.

Main Results:

• The proposed FMVCCF method demonstrates reduced sensitivity to feature dimensionality.
• The correntropy criterion and orthogonal constraints enhance factorization effectiveness and robustness.
• FMVCCF achieves promising effectiveness and robustness across various real-world datasets.

Conclusions:

• FMVCCF offers an efficient and robust solution for multi-view clustering.
• The method effectively handles high-dimensional data and complex noise.
• This approach advances the application of concept factorization in multi-view clustering.