Speaker: Harrison Hartle
Abstract:
The choice of what outcomes are possible in a model can strongly impact that model’s probabilistic structure, even under otherwise identical formulations. In models of networks, the set of possible outcomes is determined not only by any sharp constraints but also by the type of network considered (e.g., simple graphs, directed graphs, hypergraphs). We demonstrate the effects of this choice in several examples, with the primary theme being that of labeled vs unlabeled outcomes and modeling tools. We explore cases in which model predictions (and well-posedness) are impacted by sample space choice, as well as cases in which the choice has a negligible effect. We also discuss some implications for modeling of real networks.