A framework of joint graph embedding and sparse regression for dimensionality reduction.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Dimensionality reduction algorithms are crucial for data analysis.
• Existing methods often separate graph embedding and sparse regression.
• This separation limits performance due to reliance on graph construction.

Purpose of the Study:

• To propose a unified framework for joint graph embedding and sparse regression.
• To optimize embedding learning and sparse regression simultaneously.
• To enable unified supervised, semi-supervised, and unsupervised learning.

Main Methods:

• Combined objective functions of graph embedding and sparse regression.
• Joint implementation and optimization of embedding learning and sparse regression.
• Analysis of optimization problems and application of ℓ2,1-norm regularization.

Main Results:

• The proposed framework unifies various learning algorithms.
• Simultaneous feature selection and subspace learning are achieved.
• Experimental results show significant improvement in recognition performance.

Conclusions:

• Joint graph embedding and sparse regression offers superior performance over separate methods.
• The unified framework enhances feature selection and subspace learning.
• This approach provides a more effective dimensionality reduction strategy.