Limit Results for Estimation of Connectivity Matrix in Multi-layer Stochastic Block Models

Wenqing Su,Xiao Guo,Ying Yang

2024-06-17

Abstract:Multi-layer networks arise naturally in various domains including biology, finance and sociology, among others. The multi-layer stochastic block model (multi-layer SBM) is commonly used for community detection in the multi-layer networks. Most of current literature focuses on statistical consistency of community detection methods under multi-layer SBMs. However, the asymptotic distributional properties are also indispensable which play an important role in statistical inference. In this work, we aim to study the estimation and asymptotic properties of the layer-wise scaled connectivity matrices in the multi-layer SBMs. We develop a novel and efficient method to estimate the scaled connectivity matrices. Under the multi-layer SBM and its variant multi-layer degree-corrected SBM, we establish the asymptotic normality of the estimated matrices under mild conditions, which can be used for interval estimation and hypothesis testing. Simulations show the superior performance of proposed method over existing methods in two considered statistical inference tasks. We also apply the method to a real dataset and obtain interpretable results.

Statistics Theory

What problem does this paper attempt to address?

This paper attempts to address the problem of estimating and studying the asymptotic properties of scaled connectivity matrices in multi-layer stochastic block models (multi-layer SBM). Specifically, the authors focus on how to effectively estimate these connectivity matrices in multi-layer networks and establish the asymptotic normality of these estimators under mild conditions, which can be used for interval estimation and hypothesis testing.

### Background and Motivation
1. **Applications of Multi-layer Networks**: Multi-layer networks naturally arise in various fields such as biology, finance, and sociology. They can better represent multiple relationships between the same entities, enhancing the understanding of complex network data.
2. **Multi-layer Stochastic Block Model**: The multi-layer stochastic block model (multi-layer SBM) is widely used for community detection tasks in multi-layer networks. Each layer corresponds to a stochastic block model (SBM), where edges in each layer are generated according to an inter-layer block probability matrix (i.e., connectivity matrix) given unobserved communities (blocks).
3. **Limitations of Existing Research**: Most existing literature focuses on the statistical consistency of community detection methods under multi-layer SBM. However, the asymptotic distribution properties are equally important for subsequent statistical inference tasks, and this aspect has not been fully explored in multi-layer SBM.

### Main Contributions
1. **Systematic Study of Asymptotic Normality**: The authors systematically study the asymptotic normality of the estimated scaled connectivity matrices under multi-layer SBM and its variant (multi-layer degree-corrected SBM). To the best of the authors' knowledge, this is the first exploration of asymptotic normality in multi-layer degree-corrected SBM.
2. **Applicability under Mild Conditions**: Compared to previous work, this paper allows the connectivity matrix of each layer to be rank-deficient, requiring only that their sum of squares is full rank. This makes the method particularly suitable for analyzing multi-layer networks where each layer captures only part of the latent communities.
3. **Application to Statistical Inference Tasks**: The authors apply the established asymptotic normality to two statistical inference tasks: interval estimation of the scaled connectivity matrices and testing whether the overall matrices of different layers are the same. Numerical results show that the proposed method outperforms existing methods in these two tasks.

### Methods and Results
1. **Estimation Method**: The authors propose a simple and efficient method based on spectral clustering to estimate the scaled connectivity matrices. The common feature space is estimated by the eigen-decomposition of the sum of squares of the bias-adjusted adjacency matrices.
2. **Asymptotic Properties**: Under appropriate conditions, the asymptotic normality of the estimated scaled connectivity matrices under multi-layer SBM and its variant is established.
3. **Numerical Experiments**: The effectiveness of the method is validated through simulations and real data experiments. Results show that the proposed method performs superiorly in interval estimation and hypothesis testing tasks.

### Conclusion
This paper fills the gap in existing research by systematically studying the asymptotic properties of the estimated scaled connectivity matrices in multi-layer SBM and provides powerful tools for statistical inference in multi-layer networks.