Robust Semi-Supervised Subspace Clustering via Non-Negative Low-Rank Representation

Area of Science:

• Data Science
• Machine Learning
• Computer Vision

Background:

• Low-Rank Representation (LRR) effectively explores data subspace structures.
• Existing LRR methods lack robust label integration for discriminant affinity matrices.
• Independent affinity matrix construction and subspace clustering limit overall optimization.

Purpose of the Study:

• To propose a robust semi-supervised subspace clustering method using Non-Negative LRR (NNLRR).
• To explicitly incorporate supervision information for guided affinity matrix construction.
• To achieve overall optimum by integrating affinity matrix construction and subspace clustering.

Main Methods:

• Combined LRR framework with Gaussian fields and harmonic functions in a single optimization.
• Sought a non-negative low-rank matrix for sample representation.
• Imposed sparse constraints on the affinity matrix.
• Utilized an efficient linearized alternating direction method with adaptive penalty.

Main Results:

• NNLRR effectively integrates label information for improved subspace clustering.
• The method achieves overall optimum through a unified optimization process.
• Experimental results confirm NNLRR's effectiveness and robustness to noise.

Conclusions:

• NNLRR offers a robust and effective approach to semi-supervised subspace clustering.
• The integrated optimization guarantees superior performance compared to existing methods.
• NNLRR demonstrates significant advantages in handling noisy data.