Data-driven inference of hidden nodes in networks.

Area of Science:

• Complex Systems Science
• Network Inference
• Computational Neuroscience

Background:

• Complex natural systems, such as the human brain, are often not fully observable.
• Interaction network inference must account for unobserved factors influencing system behavior.

Purpose of the Study:

• To develop an effective approach for model inference in systems with hidden variables.
• To identify various interaction types (observed-to-observed, hidden-to-observed, etc.) and hidden variable configurations.

Main Methods:

• The method infers interactions and hidden variables from observed data configurations.
• Simulations using a kinetic Ising model were performed to validate the approach.
• The method was applied to real-world data from neural networks and financial markets.

Main Results:

• The proposed method outperforms existing techniques in interaction network inference.
• Predictive modeling incorporating hidden variables shows significant accuracy improvements.
• Applied to MNIST digits, the method identified approximately 60 clusters.

Conclusions:

• The developed method provides a robust framework for analyzing complex systems with unobserved components.
• Accurate inference of hidden variables enhances predictive capabilities across diverse scientific domains.
• The approach offers a novel way to determine the number of clusters in datasets, demonstrated by MNIST classification.