Title: Stratified stochastic variational inference for network factor models
Authors: Emanuele Aliverti - University Ca' Foscari of Venezia (Italy) [presenting]
Abstract: Recently, there has been considerable interest in the Bayesian modeling of networks using latent space models. As the number of nodes increases, Markov Chain Monte Carlo can be demanding, thus motivating research into alternative algorithms that scale well in high dimensions. The focus is on the latent factor model for networks and on scalable algorithms to perform approximate Bayesian inference. Leveraging sparse representations of network data and conditionally conjugate specifications, a stratified stochastic variational algorithm is developed. Empirical results demonstrate the benefit of the proposed specification in terms of computational resources and timing.