Multi-Source Data and Knowledge Fusion via Deep Learning for Dynamical Systems: Applications to Spatiotemporal...

Area of Science:

• Computational science
• Biomedical engineering
• Data science

Background:

• Advanced sensing and imaging generate vast data for spatiotemporal dynamical systems.
• Partial differential equations (PDEs) model the underlying physics of these systems.
• Integrating physics knowledge with multi-source data is crucial for accurate predictive modeling.

Purpose of the Study:

• To propose a deep learning framework for multi-source data and knowledge fusion in dynamical systems.
• To apply this framework to spatiotemporal cardiac modeling.
• To enhance the robustness and accuracy of predictive models by incorporating geometric information.

Main Methods:

• Developed a deep learning framework for fusing multi-source sensing data and physics-based knowledge.
• Incorporated physics-based information flow between different data domains.
• Utilized a graph Laplacian to integrate geometric information of 3D systems for robust modeling.

Main Results:

• The proposed framework effectively fuses multi-source data and physics knowledge.
• Demonstrated superior performance in modeling cardiac electrodynamics compared to traditional methods.
• Achieved robust spatiotemporal predictive modeling by incorporating geometric information.

Conclusions:

• The novel framework enables effective fusion of diverse data and prior knowledge for dynamical systems.
• The approach significantly improves predictive modeling accuracy, particularly for complex systems like the heart.
• This method offers a powerful tool for advancing spatiotemporal modeling in various scientific domains.