Nathan Ross, University of Melbourne

Approximating Exponential Random Graph Models
Feb 4, 2020, 4:30 pm5:30 pm
101 - Sherrerd Hall
Event Description

Network-structured data is now routinely collected and analyzed in many scientific disciplines. For example, human social networks are used to understand the effects of various interactions to the spread of disease, or to outcomes in education. Exponential random graph models (ERGMs) are widely used—especially in the sociology literature—to infer real-world network structure and function. The models are intuitively appealing, and are stationary distributions of natural Markov chains which are used to design inference procedures. On the other hand, the behavior of ERGMs is opaque, and the inference procedures can be degenerate or unstable. In this talk, I will discuss a recent thread of mathematical and statistical research that sheds light on the behavior of ERGMs, including some recent work of mine with Gesine Reinert.