Abstract:
|
Relational data have become an important component of modern statistics. Networks, and weighted networks, are ubiquitous in modern applications such as disease dynamics, ecology, financial contagion, and neuroscience. The inference of networks is harder, in parts because the model placed on the observables needs to satisfy permutation invariance, and most networks are very sparse, with most possible relations not present. This talk will explore how to best construct nonparametric summaries of such objects, in such a way that the underlying statistical model of the observations is well described by the summary, and the summary computable with scalable algorithms.
|