B0453
Title: Fast variational inference for Bayesian nonparametric latent space models for dynamic networks using Bayesian P-Splines
Authors: Joshua Loyal - Florida State University (United States) [presenting]
Abstract: Latent space models (LSMs) are often used to analyze dynamic (time-varying) networks that evolve in continuous time. However, existing approaches to Bayesian inference rely on Markov chain Monte Carlo algorithms, which cannot handle large temporal networks with many nodes or observed time points. The contributions are two-fold. First, a new prior is introduced for continuous-time LSMs based on Bayesian P-splines that are adaptive to the dimension of the latent space and the temporal smoothness of each latent position. Theoretical results are provided on the prior's flexibility by connecting it to existing Gaussian process priors. Next, a stochastic variational inference algorithm is proposed to estimate the model parameters. The approach uses stochastic optimization to sub-sample both dyads and observed time points to design a fast algorithm that scales linearly with the number of edges in the temporal network. Lastly, the model and stochastic variational inference algorithm are applied to simulated and real data to illustrate its performance.
