Identifiability and consistency of network inference using the hub model and variants: a restricted class of Bernoulli mixture models