Masked convolutional neural network for supervised learning problems.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Convolutional Neural Networks (CNNs) excel at classification and prediction but lack interpretability.
• Improving the interpretability of deep neural networks is crucial for both theoretical and practical applications.
• Existing models require enhanced methods for understanding their decision-making processes.

Purpose of the Study:

• To propose novel masked CNN (MCNN) models with improved interpretability.
• To enhance the prediction accuracy of deep learning models.
• To introduce interpretable representations within neural network architectures.

Main Methods:

• Introduction of a latent binary network within the CNN architecture.
• Extraction of informative regions of interest (ROIs) using the latent binary network.
• Integration of the latent binary network with CNNs for supervised learning tasks.

Main Results:

• The proposed MCNN models demonstrate superior interpretability compared to traditional CNNs.
• MCNNs achieve competitive prediction accuracy in various supervised learning problems.
• Numerical studies validate the effectiveness of the integrated latent binary network approach.

Conclusions:

• Masked CNNs offer a promising solution for enhancing deep learning model interpretability.
• The integration of latent binary networks effectively extracts critical information for improved predictions.
• MCNNs represent a significant advancement in developing more transparent and accurate neural networks.