10-601 Network Models
From Cohen CoursesJump to navigationJump to search
E. Airoldi, D. Blei, S. Fienberg, and E. P. Xing,Mixed Membership Stochastic Blockmodel Journal of Machine Learning Research, 9(Sep):1981--2014, 2008.
Taking home message
- Networks can be described by global/local features (what are they?)
- But these features are not helpful for inferring individual nodal information.
- SBM is a probabilistic model for grouping network nodes into communities/clusters, each cluster corresponds to a single "role" or social position
- MMSB is a probabilistic model that allows each node to have multiple roles, and allow each edge to have contextual-dependent role instantiations.
- dMMSB can infer multi-role trajectories for every node.
- Inference of MMSB and dMMSB employs MCMC or VI