First seminar

Speaker: Michael Fop

Title: Bayesian nonparametric latent position models for complex networks

Abstract:
Latent position models provide a probabilistic and geometric framework for analyzing network data, representing nodes as points in a low-dimensional space where proximity reflects tie likelihood. This enables interpretable visualization and model-based clustering of complex relational structures across social, biological, and spatial systems. Traditional formulations often require pre-specifying the number of clusters and latent dimensions or rely on expensive model selection, limiting flexibility.
This talk presents recent advances in Bayesian nonparametric latent position models that jointly infer latent dimensionality and clustering while quantifying uncertainty. Within a unified probabilistic framework, the models extend naturally to multidimensional, multilayer, temporal, and spatial settings, balancing complexity and interpretability. Applications span social interaction networks, neuroscientific connectivity, and spatio-temporal interaction systems, showing how Bayesian nonparametrics reveal meaningful latent geometry and community organization.

Second seminar

Speaker: Alberto Caimo

Title: Multi-layer dissolution exponential-family models for weighted signed networks

Abstract:

Understanding the structure of weighted signed networks is crucial for analysing social systems where relationships differ in both sign and intensity. Although statistical network analysis has advanced considerably, there remains a shortage of models capable of simultaneously and rigorously capturing both the sign and strength of edges. To address this gap, we propose a multi-layer dissolution exponential random graph modelling framework that jointly represents the signed and weighted processes, conditional on the observed interaction structure. This framework allows for a principled evaluation of structural balance effects while fully incorporating edge weights. To strengthen inference, we employ a fully probabilistic Bayesian hierarchical approach that partially pools information across layers, with parameters estimated via an adaptive approximate exchange algorithm. We illustrate the flexibility and explanatory power of our methodology using bill sponsorship data from the 108th US Senate, uncovering complex patterns of signed and weighted interactions as well as structural balance effects that traditional approaches cannot capture.