Rank- 2-D Multinomial Logistic Regression for Matrix Data Classification.

Area of Science:

• Machine Learning
• Data Science
• Pattern Recognition

Background:

• The rapid growth of matrix data necessitates efficient classification methods.
• Existing 2-D methods like linear discriminant analysis and logistic regression have limitations.

Purpose of the Study:

• To propose a novel 2-D framework, rank-2D multinomial logistic regression (2DMLR-RK), for multiclass matrix classification.
• To address the limitations of current 1-D and 2-D classification techniques.

Main Methods:

• The 2DMLR-RK models each category using rank-constrained left and right projection matrices.
• Left projection matrices capture row information; right projection matrices capture column information.
• An effective framework is developed to solve the non-convex optimization problem, considering convergence, initialization, and computational complexity.

Main Results:

• The parameter balances learning capacity and generalization in 2DMLR-RK.
• Extensive experiments demonstrate superior performance on diverse datasets.
• 2DMLR-RK achieves higher classification accuracy and reduced computation time compared to 1-D methods.

Conclusions:

• 2DMLR-RK outperforms existing 1-D and state-of-the-art 2-D methods for matrix data classification.
• The proposed method offers a robust and efficient solution for multiclass matrix classification problems.