Student seminar: Peter MacDonald

Peter MacDonald
Assistant Professor, University of Waterloo

Room: M3 3127

Latent space models for network data: an overview

This talk will introduce latent space models for complex relational (network) data. Latent space models enable interpretability, dimension reduction, theoretical tractability, and naturally adapt to missing data. Canonical formulations such as the random dot product graph, and estimation strategies including likelihood-based, spectral, and machine learning methods will be reviewed, including an overview of asymptotic theory for the random dot product graph. Selected research directions of interest, including multilayer and dynamic networks, alternative geometries, and applications to regression and causal inference will be discussed.