🤖 AI Summary
This paper addresses unsupervised multiple changepoint detection in dynamic network time series. We propose a changepoint localization method based on the Separable Temporal Exponential Random Graph Model (STERGM). Methodologically, we introduce the first coupling of the Alternating Direction Method of Multipliers (ADMM) with Group Fused Lasso (GFL) regularization to jointly estimate abrupt changes in time-varying STERGM parameters. To enhance model selection stability, we design a STERGM-adapted modified Bayesian Information Criterion (BIC). Our approach achieves significantly improved changepoint localization accuracy on both synthetic and real-world network datasets, enables simultaneous detection of multiple changepoints, and supports interpretable inference of underlying generative mechanisms. The proposed algorithms are publicly available as the R package CPDstergm.
📝 Abstract
This paper studies the unsupervised change point detection problem in time series of networks using the Separable Temporal Exponential-family Random Graph Model (STERGM). Inherently, dynamic network patterns can be complex due to dyadic and temporal dependence, and change points detection can identify the discrepancies in the underlying data generating processes to facilitate downstream analysis. Moreover, the STERGM that utilizes network statistics to represent the structural patterns is a flexible and parsimonious model to fit dynamic networks. We propose a new estimator derived from the Alternating Direction Method of Multipliers (ADMM) procedure and Group Fused Lasso (GFL) regularization to simultaneously detect multiple time points, where the parameters of a time-heterogeneous STERGM have changed. We also provide a Bayesian information criterion for model selection and an R package CPDstergm to implement the proposed method. Experiments on simulated and real data show good performance of the proposed framework.
