Title: Statistical modeling of sparse networks

Abstract: The study of scientific and social phenomena often requires modeling data in the form of networks, i.e. the set of interactions between entities like proteins, neurons or people. Statistical models provide ways to address important questions, including why some entities are connected but not others, and whether there are interactions that have not been observed. In this talk, I will present my work on developing a class of models that can discover an unobserved set of clusters (or communities) among interacting entities and that can learn the number of clusters from data. Unlike previous approaches with those properties, the proposed framework can model sparse networks. Capturing sparsity is especially important when dealing with large networks: for example, in online social networks, someone’s connections grow much slower than linearly with the number of users. I will summarize a posterior-inference method based on Markov Chain Monte Carlo and I will show the advantages of this approach on a set of social and biological networks.