Bayesian Nonparametric Modelling for Sparse Dynamic Graphs With Reciprocating Relationships
Bayesian statistics and machine learning have shown great advances over the last few years providing us with powerful approaches to model data generating mechanisms and quantify the uncertainty therein. Probabilistic models with bayesian nonparametric assumptions have been used in many applications of machine learning such as density estimation, clustering, latent feature models and survival analysis. In this talk I will focus on probabilistic modelling for networks and temporal interaction data. The suggested class of models is based on point processes and captures both the global graph features such as sparsity, community structure and power-law degree as well as the local dynamics within each pair such as reciprocating relationships. This network model family is highly expressive, interpretable, unifies several graph classes and outperforms alternative models in link prediction.
See here for more information about the Rough Path Interest Group.