End-to-End Ensemble Learning by Exploiting the Correlation Between Individuals and Weights.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Computer Science

Background:

• Ensemble learning enhances performance through diverse classifiers.
• Balancing classifier diversity and individual accuracy is key for optimal ensemble results.

Purpose of the Study:

• Propose a novel ensemble method optimizing the trade-off between classifier diversity and accuracy.
• Develop a joint optimization model to enhance ensemble classification performance.

Main Methods:

• Constructing a joint optimization model to exploit correlations between classifiers and weights.
• Modeling the framework as a shallow network for efficient end-to-end training.
• Imposing sparsity constraints on weights for selective classifier usage.

Main Results:

• Achieving high total classification performance via weighted classifiers.
• Enabling individual classifier updates based on the ensemble's error.
• Demonstrating improved classification performance on UCI datasets compared to existing methods.

Conclusions:

• The proposed ensemble method effectively balances diversity and accuracy.
• The framework offers an efficient and adaptable approach to ensemble learning.
• The method shows significant improvements in classification tasks.