Learning Coupled System Dynamics with Incomplete Information
Research Paper#Physics-Informed Machine Learning, Coupled Systems, Neural Networks🔬 Research|Analyzed: Jan 3, 2026 19:14•
Published: Dec 28, 2025 22:02
•1 min read
•ArXivAnalysis
This paper addresses a significant challenge in physics-informed machine learning: modeling coupled systems where governing equations are incomplete and data is missing for some variables. The proposed MUSIC framework offers a novel approach by integrating partial physical constraints with data-driven learning, using sparsity regularization and mesh-free sampling to improve efficiency and accuracy. The ability to handle data-scarce and noisy conditions is a key advantage.
Key Takeaways
- •Addresses the problem of modeling coupled systems with incomplete physics and missing data.
- •Introduces MUSIC, a sparsity-induced multitask neural network framework.
- •Employs mesh-free sampling and sparsity regularization for efficiency.
- •Demonstrates accurate learning of solutions under data-scarce and noisy conditions.
- •Outperforms non-sparse formulations in experiments.
Reference / Citation
View Original"MUSIC accurately learns solutions to complex coupled systems under data-scarce and noisy conditions, consistently outperforming non-sparse formulations."