ABSTRACT
This chapter discusses a generic class of relatively exchangeable network models which incorporates population heterogeneity into the graphon framework. It explains the stochastic blockmodel (SBM) in order to account for more complex heterogeneity in network data. The conditional independence structure of latent space models (LSMs), and relatively exchangeable random graphs more generally, prevents the model from accounting for higher-order structure, such as relationships among three or more vertices. SBMs and LSMs are all special kinds of relatively exchangeable network models. Before discussing about relative exchangeability, the chapter proposes the refinement of the Caron-Fox model. It presents relatively invariant point process models as a wide open topic for future applied and theoretical research. The chapter emphasizes the basic ideas underlying relative exchangeability, but as of now the available results involve obscure mathematical notions and difficult techniques which lie beyond the scope of a typical discussion on statistical network analysis.
