Neural-prior stochastic block model

Abstract The stochastic block model (SBM) is widely studied as a benchmark for graph clustering aka community detection. In practice, data often come with node attributes that bear additional information about the communities. Previous works modeled such by considering are generated from memberships. this work, motivated recent surge of in signal processing using deep neural networks priors, we propose to communities being determined rather than opposite. We define corresponding model; call it neural-prior SBM. an algorithm, stemming statistical physics, based on combination belief propagation and approximate message passing. analyze performance algorithm well Bayes-optimal performance. identify detectability exact recovery phase transitions, algorithmically hard region. proposed can be used both theory algorithms. To illustrate this, compare optimal performances simple networks.