Structure Mapping Generative Adversarial Network for Multi-View Information Mapping Pattern Mining.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Multi-view learning aims to integrate diverse data sources for improved model generalization.
• Existing methods often treat different data views independently, neglecting shared information mapping patterns.
• This limits the potential for synergistic learning and deeper insights from complementary data.

Purpose of the Study:

• To propose a novel Structure Mapping Generative adversarial network (SM-GAN) framework.
• To exploit the consistency and complementarity of multi-view data through information mapping.
• To capture hierarchical interaction patterns within network-structured multi-view data.

Main Methods:

• Developed a structural information mapping model for network-structured multi-view data.
• Integrated three types of graph convolutional operations within the SM-GAN generator.
• Implemented a structural information mapping module between the encoder and decoder for micro- to macro-view mapping.

Main Results:

• SM-GAN demonstrated superior performance compared to baseline and advanced methods.
• Experiments were validated using imaging genetics data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset.
• The framework effectively addressed multi-label classification and evolution prediction tasks.

Conclusions:

• The proposed SM-GAN framework successfully utilizes structural information mapping for multi-view learning.
• This approach enhances the integration of information from different views, leading to improved model performance.
• SM-GAN offers a promising direction for analyzing complex, multi-modal datasets in neuroscience and beyond.