Dual Low-Rank Decompositions for Robust Cross-View Learning.

Area of Science:

• Computer Science
• Machine Learning
• Data Science

Background:

• Cross-view data, essential for rich data representation, suffers from significant divergence.
• Data from the same category across different views exhibit lower similarity than data within the same view but different categories.

Purpose of the Study:

• To propose a robust cross-view learning framework.
• To seek a robust view-invariant low-dimensional space.
• To mitigate view divergence and capture within-class knowledge.

Main Methods:

• Developed a dual low-rank decomposition technique to separate intertwined class and view manifold structures.
• Designed two discriminative graphs to constrain the decomposition using prior knowledge.
• Created a flexible framework for scenarios with unknown view information during evaluation.

Main Results:

• The proposed algorithm effectively captures within-class knowledge.
• Achieved mitigation of view divergence, leading to a more effective view-invariant feature extractor.
• Demonstrated superior performance on face and object benchmarks compared to state-of-the-art methods.

Conclusions:

• The dual low-rank decomposition effectively addresses cross-view data divergence.
• The framework provides a flexible and effective solution for cross-view learning, even with incomplete view information.
• The method significantly enhances feature extraction for cross-view recognition tasks.