Broad Learning With Reinforcement Learning Signal Feedback: Theory and Applications.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Broad learning systems (BLSs) are recognized for efficient discriminative learning.
• Standard BLS methods have limitations in performance improvement and autonomous structural optimization.

Purpose of the Study:

• To propose a modified Broad Learning System with Reinforcement Learning Feedback (BLRLF).
• To enhance the performance and adaptability of standard BLS through novel optimization techniques.

Main Methods:

• Implemented weight optimization using value iteration (VI)-based adaptive dynamic programming (ADP) for connection weight increments.
• Integrated broad expansion methods with a heuristic search for autonomous network structure optimization.
• Utilized reinforcement learning signal feedback for adaptive learning.

Main Results:

• BLRLF demonstrates superior performance compared to standard BLS and other state-of-the-art algorithms.
• The proposed method achieves high accuracy on benchmark datasets from the UC Irvine Machine Learning Repository and other challenging datasets.
• BLRLF maintains a fast computational nature despite increased training time.

Conclusions:

• BLRLF offers an effective approach to improve the performance of Broad Learning Systems.
• The autonomous structural optimization and weight adaptation provide significant advantages over existing methods.
• BLRLF represents a promising advancement in efficient and high-performing machine learning models.