A1113
Title: Nonparametric estimation of time-varying graphon model by double-smoothing
Authors: Jeonghwan Lee - University of Minnesota, Twin Cities (United States) [presenting]
Abstract: Dynamic networks capture how connections among interacting entities evolve and arise in fields as diverse as neuroscience, international economics, and social communication. Estimating the time-varying edge probability matrix from a single observed sequence of adjacency matrices is ill-posed unless additional structure is imposed. The dynamic graphon paradigm is adopted, positing that edge probabilities vary smoothly in both latent node positions and time, and a computationally tractable double-smoothing estimator that exploits this two-way regularity is devised. The procedure first smooths across time via local polynomial regression, then smooths across nodes via neighborhood smoothing, delivering non-parametric estimates of the edge probability matrix without requiring community labels or low-rank constraints. Simulation studies demonstrate accurate recovery of evolving probabilities and sharp detection of structural breaks. The proposed framework extends dynamic graphon methodology, bridges estimation, and inference in evolving networks, and offers a practical tool for researchers analyzing large, noisy relational time series.
