Factor Graph Fragments for Measurement Error and Missing Data Models


Factor graph for a logistic additive model.

Recently, Chief investigator Wand and his has group published Kim & Wand (2016, Electronic Journal of Statistics), Wand (2017, Journal of American Statistical Association) and Nolan & Wand (2017, Stat) on message passing approaches to semiparametric regression analysis. These articles advocate a new-wave general approach to statistical analyses in the face of big data sets and models. Wand (2017) introduced the notion of factor graph fragments as a means of compartmentalising requisite algebra and computer code. The project will build on this work and develop factor graph fragment updates, based on both variational message passing and expectation propagation principles, for semiparametric regression model components corresponding to data that are subject to measurement error or are partially missing.

Project Researchers

Lead CI

Green Acorn