Explainable machine learning by SEE-Net: closing the gap between interpretable models and DNNs.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Deep Neural Networks (DNNs) excel in accuracy but are often
• black-box
• models, hindering prediction interpretability.
• Interpretable statistical models typically exhibit lower accuracy compared to DNNs.

Purpose of the Study:

• To develop a novel neural network architecture that combines the high accuracy of DNNs with the interpretability of simpler models.
• To address the trade-off between prediction accuracy and model explainability in machine learning.

Main Methods:

• Introduction of the Synced Explanation-Enhanced Neural Network (SEE-Net) architecture.
• SEE-Net integrates a guiding DNN (black-box) with a shallow neural network (white-box) for co-supervision.
• The shallow network is trained under the guidance of the DNN to ensure explainability.

Main Results:

• SEE-Net successfully bridges the gap between deep learning's predictive power and the need for explainable models.
• Experiments on image and tabular data validated SEE-Net's ability to provide interpretable predictions.
• The architecture achieves high explainability with minimal compromise on prediction accuracy.

Conclusions:

• SEE-Net represents a new paradigm in machine learning, offering a practical solution for interpretable AI.
• The model's co-supervision approach allows for tailored applications requiring both performance and transparency.
• This work suggests that inadequate training, not inherent limitations, causes underperformance in some interpretable models.