Weighted Generalized Cross-Validation-Based Regularization for Broad Learning System.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Deep Learning

Background:

• The broad learning system (BLS) is a powerful flat network for classification and regression.
• Regularization is crucial for BLS performance, but fixed parameters hinder dynamically expanded networks.
• Dynamic network expansion often leads to decreased classification accuracy with predetermined regularization.

Purpose of the Study:

• To address the performance degradation of broad learning systems (BLS) with fixed regularization during dynamic expansion.
• To propose an automated method for selecting optimal regularization parameters tailored to specific datasets.
• To improve the robustness and accuracy of BLS, particularly in incremental learning scenarios.

Main Methods:

• Implementation of a novel method based on weighted generalized cross-validation (WGCV).
• Automated selection of regularization parameters for broad learning systems (BLS).
• Experimental validation on various datasets to assess performance improvements.

Main Results:

• The proposed WGCV method significantly enhances the performance of the broad learning system (BLS).
• The WGCV approach effectively alleviates the accuracy decrease typically observed in incremental learning with BLS.
• Demonstrated improved classification and regression capabilities of BLS through automated regularization.

Conclusions:

• Weighted generalized cross-validation (WGCV) provides an effective solution for optimizing regularization in broad learning systems (BLS).
• The method ensures consistent performance of BLS even when the network architecture is dynamically expanded.
• This research contributes to more reliable and accurate applications of broad learning systems in machine learning.