Overview
Matthew Stephens is a statistician (University of Oxford at the time of this paper; later University of Chicago) working on Bayesian statistics, statistical genetics, and computational methods. He is known for foundational work on mixture models and for widely used methods in population genetics (e.g. PHASE haplotype inference).
Key Contributions / Features
- Label switching — Stephens (2000): The standard decision-theoretic relabelling-algorithm solution to the label-switching problem in Bayesian mixture models, showing identifiability constraints fail in general and minimizing a Kullback–Leibler loss on clusterings instead. See Stephens (2000) and Mixture Model.
- Mixtures with an unknown number of components (2000, Annals of Statistics): A birth-death MCMC alternative to reversible-jump methods for mixtures of unknown dimension.
Related