Nonparametric Bayesian inference of the microcanonical stochastic block model

Area of Science:

• Network Science
• Statistical Physics
• Computer Science

Background:

• Characterizing hidden network structures is crucial for understanding complex systems.
• Generative models and parameter inference are key approaches.
• The stochastic block model (SBM) is suitable for modular network structures.

Purpose of the Study:

• To present a nonparametric Bayesian method for inferring modular network structure.
• To determine the number of modules and their hierarchical organization.
• To improve upon traditional inference approaches for empirical networks.

Main Methods:

• Utilizing a microcanonical variant of the SBM with hard constraints.
• Implementing deeper Bayesian hierarchies with sequences of priors and hyperpriors.
• Developing an efficient inference algorithm scalable to large networks and numerous modules.

Main Results:

• The method successfully infers the number and hierarchical organization of network modules.
• Deeper Bayesian hierarchies overcome limitations of traditional methods on large networks.
• The efficient algorithm scales well for large networks and an unlimited number of modules.
• The approach enables sampling modular hierarchies and performing model selection.

Conclusions:

• The proposed nonparametric Bayesian method offers significant improvements for network structure inference.
• It provides a scalable and effective tool for analyzing complex modular networks.
• The method allows for robust model selection and posterior sampling of network hierarchies.