Title: Subsampling and inference for beta neutral-to-the-left models of random graphs
Authors: Benjamin Bloem-Reddy - University of Oxford (United Kingdom) [presenting]
Abstract: Beta Neutral-to-the-Left (NTL) models are able to generate random graphs with any level of sparsity, and with power law degree distributions of any possible exponent. This flexibility is unique among models in the recent literature based on various notions of exchangeability, which are able to obtain sparsity values and power law exponents over a limited range of possible values. The flexibility of Beta NTL models comes at the cost of losing any obvious form of exchangeability. However, by conditioning on (or inferring) the vertex arrival times, a constrained notion of exchangeability becomes apparent, and may be exploited for efficient estimation and inference algorithms, as well as for model-coherent subsampling, even when no ordering information is available.