Structured Sparse Subspace Clustering: A Joint Affinity Learning and Subspace Clustering Framework.

Area of Science:

• Data Mining
• Machine Learning
• Computer Vision

Background:

• Subspace clustering segments data from multiple subspaces.
• Current methods use a two-stage approach: affinity matrix learning then spectral clustering.
• This two-stage method is suboptimal as affinity and segmentation are interdependent.

Purpose of the Study:

• To propose a joint optimization framework for subspace clustering.
• To develop Structured Sparse Subspace Clustering (S³C) for simultaneous affinity and segmentation learning.
• To extend S³C to Constrained S³C (CS³C) incorporating side-information.

Main Methods:

• Proposing the S³C framework using structured sparse representations.
• Developing CS³C to integrate partial side-information during affinity learning.
• Employing alternating direction method of multipliers with spectral clustering for optimization.

Main Results:

• Demonstrated effectiveness of S³C and CS³C on synthetic and real-world datasets.
• Achieved state-of-the-art results in subspace clustering tasks.
• Validated performance on face recognition, motion segmentation, and cancer data.

Conclusions:

• The joint optimization framework significantly improves subspace clustering.
• S³C and CS³C offer a more effective approach by exploiting interdependencies.
• The methods show broad applicability across diverse data types and applications.