Active Surveillance via Group Sparse Bayesian Learning.

Area of Science:

• Complex Systems
• Computational Science
• Data Science

Background:

• Effective control of diffusion systems requires accurate dynamic prediction.
• Real-world resource constraints limit comprehensive system observation.
• Predicting large spatio-temporal diffusion dynamics from partial observations is a key challenge.

Purpose of the Study:

• Develop a novel computational method for active surveillance with limited resources.
• Predict diffusion system dynamics using observations of a subset of components.
• Identify key components within a diffusion system for efficient monitoring.

Main Methods:

• Introduced a novel measure, the gamma (γ) value, to identify critical components.
• Modeled a sentinel network with a row sparsity structure.
• Designed a backward-selection sentinel network mining algorithm (SNMA) using group sparse Bayesian learning.
• Addressed computational scalability and extended SNMA to non-linear dynamical systems.

Main Results:

• The sentinel network mining algorithm (SNMA) effectively identifies key components in diffusion systems.
• SNMA demonstrates scalability for large spatio-temporal systems.
• The method is effective for both linear and non-linear diffusion dynamics.
• Validation on synthetic and real-world datasets shows SNMA outperforms state-of-the-art methods.

Conclusions:

• Active surveillance using SNMA provides an effective approach for predicting diffusion system dynamics under resource constraints.
• The γ value and SNMA offer a robust framework for identifying critical system components.
• SNMA represents a significant advancement in the field, offering practical solutions for complex diffusion system monitoring.