A0419
Title: Bayesian covariate-dependent latent space model with information adaptivity
Authors: Peng Zhao - University of Delaware (United States) [presenting]
Yabo Niu - University of Houston (United States)
Abstract: Modern network data analysis often involves analyzing network structures alongside covariate features to gain deeper insights into underlying patterns. However, traditional statistical network models may not fully consider the integration of such rich node characteristics. To address this gap, a new Bayesian high-dimensional covariate-dependent latent space model is introduced. This framework links latent vectors representing network structures with low-rank approximations of high-dimensional covariate observations, capturing their mutual dependencies. To adaptively integrate dependencies, a shrinkage is used prior to the discrepancy between latent network vectors and low-rank covariate approximation vectors, accommodating both consistencies and inconsistencies between them. To achieve computation efficiency, a mean-field variational inference algorithm is developed to approximate the posterior distribution. The concentration rate of the posterior is established within a suitable parameter space, and it is demonstrated how the model facilitates adaptive information aggregation between networks and covariates. Extensive simulations and real-world data analyses confirm the effectiveness of our approach.
