Semisupervised Regression With Optimized Rank for Matrix Data Classification.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Existing vector-based classification methods often disregard matrix structural information.
• Vectorization can lead to high dimensionality, causing overfitting with limited training data.

Purpose of the Study:

• To develop a matrix-based regression algorithm for classification that directly uses matrix data.
• To overcome the limitations of traditional vectorization techniques in matrix data classification.

Main Methods:

• A novel matrix-based regression algorithm is proposed, directly learning regression matrices from input matrices.
• A joint l2,1-norm is incorporated to optimize regression rank by identifying common sparse columns.
• A semi-supervised learning process is integrated, leveraging both labeled and unlabeled data.

Main Results:

• The proposed method effectively utilizes the inherent structure of matrix data.
• Experiments demonstrate superior performance compared to state-of-the-art methods on benchmark datasets.
• The algorithm shows strong classification performance even with a small number of labeled training samples.

Conclusions:

• The matrix-based regression approach offers a significant improvement over vector-based methods for matrix data classification.
• The integration of joint sparsity and semi-supervised learning enhances discrimination and classification accuracy.
• This method provides an effective solution for classification tasks with limited labeled data.