Focus
Our lab develops Bayesian methods for settings where data are scarce, noisy, high-dimensional, spatially structured, or collected under changing experimental conditions.
Our goal is to build models that provide accurate inference and prediction while communicating uncertainty and supporting reproducible data analysis.
We develop flexible nonparametric Bayesian theory, including Dirichlet-process extensions and dependent species sampling models.
Methods include Gaussian process regression and classification, spectral analysis models (bsamGP), and Bayesian deep kernel machines.
We study shrinkage priors for variable selection and dimension reduction in high-dimensional statistical settings.
Our work includes sparse covariance and precision-matrix estimation with screened beta-mixture shrinkage priors (bspcov), as well as Horseshoe and spike-and-slab priors.
We develop efficient Bayesian algorithms that address the computational demands of large and high-dimensional data.
Key methods include Gaussian variational approximation (vbayesGP), Pólya–Gamma augmentation, and accelerated Bayesian kernel machine regression.
We apply Bayesian modeling to molecular safety and material properties; biomedical, health, and neuroimaging studies; and environmental, agricultural, fisheries, and spatial data.
Representative applications include molecular toxicity and physicochemical-property prediction, dynamic fMRI analysis, radiation-response modeling, aquaculture growth prediction, weather forecasting, wildfire risk, and satellite imagery.
