|
Abstract:
|
The increasing prevalence of high-dimensional longitudinal network data motivates the need for efficient and reliable inference schemes. In this context, high-dimensional may refer to either a network with a large number of nodes, a network for which we have many observations through time, or both. Although there are a range of network modelling approaches, we restrict our attention to the Dynamic Latent Space Model of Sewell and Chen (2016). For this modelling approach inference is typically carried out through either MCMC schemes or variational methods. We instead note that this model presents a natural framework for Sequential Monte Carlo (SMC) techniques, whereby we sequentially estimate the latent network structure for each observation in time. Furthermore, SMC algorithms are amenable to online updating and hence allow the exploration of online inference for streaming networks.
|
