|
Abstract:
|
The onslaught of network datasets from disciplines as diverse as neuroscience, finance, telecommunications, cybersecurity, and social science has created a major opportunity for growth in our field. These opportunities bring with them computational, logistical, and conceptual challenges, which are compounded by the shortage of suitable models for modern network data. Apart from the difficulty of accounting for complex structural properties in these datasets (e.g., sparsity, clustering, heterogeneity, etc.), the development of sound models is further confounded by a poor understanding of how network sampling and dynamics affect inferences. I'll cover several examples that are designed to illustrate how these various considerations factor into model specification for networks. These examples touch on the strengths and weaknesses of exponential random graph models, stochastic blockmodels, graphon models, edge exchangeable models, and dynamic network models.
|
