Attribute fusion in a latent process model for time series of graphs

We consider anomaly/change point detection given a time series of graphs with categorical attributes on the edges. Various attributed graph invariants are considered, and their power for detection as a function of a linear fusion parameter is presented.