Groupwise Label Enhancement Broad Learning System for Image Classification.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Neural Networks

Background:

• The broad learning system (BLS) offers efficient learning but is constrained by binary labels.
• Current label enhancement methods can reduce intraclass similarity, hindering classification performance.
• Maintaining intraclass similarity is crucial for effective classification tasks.

Purpose of the Study:

• To propose a groupwise label enhancement BLS model that preserves intraclass similarity and enhances interclass disparity.
• To develop a novel regression target and groupwise constraint for improved label representation.
• To ensure computational efficiency and theoretical convergence through an advanced optimization algorithm.

Main Methods:

• Developed a novel regression target generalizing existing BLS enhancement methods.
• Introduced a groupwise constraint to simultaneously optimize intraclass similarity and interclass disparity.
• Implemented an alternating direction method of multipliers (ADMM) for efficient model optimization.

Main Results:

• The proposed groupwise label enhancement BLS model effectively maintains intraclass similarity.
• The model successfully increases interclass disparity, improving classification accuracy.
• Experimental results show superior effectiveness and efficiency compared to state-of-the-art methods on public datasets.

Conclusions:

• The groupwise label enhancement BLS model addresses the limitations of binary labels.
• The novel approach achieves a balance between intraclass similarity and interclass disparity.
• The proposed method offers a computationally efficient and theoretically convergent solution for classification tasks.