Bayesian nonparametric models for prediction in networks

Why model networks? Many datasets take the form of networks or graphs... Social networks have binary (is friend, follows) or integer (retweets, shares) edges. Email networks have integer (number of emails) edges. Biological networks may have binary, integer or real-valued edges. Sinead Williamson Nonparametric networks 2 / 25 Modeling networks: Prediction vs Description There are a number of reasons we might want to model networks... Network recovery: We may only have a noisy version of the underlying network. Description/characterization: We may wish to find a latent explanation of the network structure – e.g. community detection. Anomaly detection: We may wish to detect unusual nodes or sub-graphs- for example to detect spammers. Influence modeling: We may wish to identify “trend-setters ” in a social network