Supervised Deep Multimodal Matrix Factorization for Interpretable Brain Network Analysis

Researchers introduced SD3MF, a supervised encoder-decoder model for multimodal brain graph analysis.

The method extends symmetric nonnegative matrix tri-factorization to learn shared latent structure and modality-specific factorization.

On connectome datasets, SD3MF outperformed CNN and GNN baselines in prediction and graph reconstruction.

Its key advantage is combining strong accuracy with interpretable brain connectivity features for neuroscience research.