Flexible models that let the data determine their own complexity — spanning Bayesian nonparametrics, variational inference, and mixture modeling.
Back to ResearchMy work in Bayesian nonparametrics develops models that let the data determine their own complexity, rather than fixing the number of components or a functional form in advance. A central thread is Bayesian nonparametric latent class analysis and Dirichlet process mixture modeling, with applications to clustering, density estimation, and missing data.
I am also interested in scalable estimation for these models through variational inference, and in making nonparametric methods practical and accessible for applied researchers.