A0657
Title: Model-based clustering of population of networks via extended stochastic block models
Authors: Maria Francesca Marino - University of Florence (Italy) [presenting]
Monia Lupparelli - University of Florence (Italy)
Giulia Capitoli - University of Milano-Bicocca (Italy)
Abstract: The population of networks arises when interactions between nodes are observed repeatedly over a set of units. The stochastic block model is extended to provide a joint clustering of units and network nodes. This is done by considering in the model specification a set of unit- and node-specific, discrete, latent variables, able to capture dependences among the observed data. A former set of latent variables allows partition units into classes based on the observed, unit-specific, network features. A latter set of latent variables is employed to identify clusters of stochastically equivalent nodes sharing similar connectivity profiles. Parameter estimation is conducted within a maximum likelihood framework. However, deriving the likelihood function is a challenging task, as it would require the solution of a multiple summation over all latent variables in the model. To solve the issue, the use of appropriate approximation methods is required. The effectiveness of the proposal is evaluated both via simulations and a real data application from the medical field.
