Dynamic Network Model Accurately Identifies Abrupt Structural Changes

A research team has introduced a fused lasso exponential random graph model designed to estimate time-varying parameters across sequences of observed dynamic networks. The new model, termed FL-ERGM, specifically addresses network ensembles that experience sudden structural shifts at particular time points while remaining stable or undergoing minimal variation between those events.

Sudden structural changes interspersed with periods of stability are characteristic of diverse real-world systems, including international trade networks and political co-voting networks. To uncover these underlying network-generating mechanisms, the proposed methodology penalizes parameter differences between consecutive time points, encouraging parameter constancy during stable intervals while permitting abrupt adjustments when supported by empirical observations.

To overcome the computational intractability of evaluating the exact exponential random graph model likelihood, the authors adopted a pseudo-likelihood approximation. They solved the resulting penalized objective function using the alternating direction method of multipliers and mathematically proved that the solution delivered by this algorithm matches the optimal solution of the original objective.

In Monte Carlo simulation experiments, the FL-ERGM approach outperformed existing methods in estimating time-varying model parameters and identifying structural change points. When evaluated on empirical data from international trade and political co-voting networks, the framework demonstrated interpretable and contextually relevant patterns of temporal parameter variation.

Subject of Research: Time-varying parameter estimation in dynamic exponential random graph models

Article Title: Dynamic exponential random graph models with the fused lasso penalty

Article References: Ren, S., Liu, P., Zhang, H., Wang, X., & Zhang, J. (2026). Dynamic exponential random graph models with the fused lasso penalty. Applied Network Science. https://doi.org/10.1007/s41109-026-00846-8

Image Credits: AI Generated

DOI: 10.1007/s41109-026-00846-8

Keywords: exponential random graph models, dynamic networks, fused lasso, structural change detection, pseudo-likelihood approximation, ADMM algorithm, network analysis, Monte Carlo simulation, Dynamic, exponential, random, graph