Mixing times of data-augmentation Gibbs samplers for high-dimensional probit regression
Researchers have published a paper on arXiv detailing the mixing times of data-augmentation Gibbs samplers used in Bayesian probit regression. The study provides explicit non-asymptotic bounds on these mixing times, which are dependent on the design matrix and prior precision. The findings identify scenarios where mixing times remain bounded even as the number of data points and parameters increase, offering guidance on selecting prior distributions for faster convergence. An empirical analysis using coupling techniques supports the effectiveness of these bounds in predicting practical behaviors. AI