Multigraph classification using learnable integration network with application to gender fingerprinting.

Area of Science:

• Graph Neural Networks
• Machine Learning
• Computational Neuroscience

Background:

• Multigraphs with heterogeneous views pose challenges for classification.
• Existing feature selection methods fail to preserve topological properties and suffer from cumulative errors.

Purpose of the Study:

• To introduce MICNet, the first end-to-end graph neural network for multigraph classification.
• To address limitations of existing methods by preserving topological properties and enabling end-to-end learning.

Main Methods:

• Developed MICNet, integrating heterogeneous multigraph views into a single representation using a GNN.
• Employed a geometric deep learning block for classification of the integrated graph template.
• Trained the model end-to-end with a single objective function.

Main Results:

• MICNet successfully integrated heterogeneous multigraph views while preserving topological properties.
• The model achieved superior performance in gender classification using brain multigraphs.
• MICNet significantly outperformed its variants, demonstrating its potential.

Conclusions:

• MICNet offers a powerful end-to-end solution for multigraph classification.
• The model effectively handles heterogeneity in multigraphs by preserving crucial topological information.
• MICNet shows significant promise for applications in complex graph analysis and classification.