Overview
Chuanhai Liu is a statistician at Bell Laboratories, Lucent Technologies. His research focuses on Bayesian computation, EM-type algorithms and their accelerated variants, and robust statistical modeling with t-distributions. He is known for the PX-EM algorithm (Liu-Rubin-Wu 1998), the CA-DA (covariance-adjusted data augmentation) algorithm (Liu 1999), and the robit regression model (Liu 2006).
Key Contributions
- Robit Regression (Liu 2006): Proposed the robit model — binary regression with a t-link — and proved the MLE has a bounded influence function; derived EM, ECME, and PX-EM algorithms for MLE and E-DA 1/E-DA 2 algorithms for Bayesian inference, with the accelerated DA schemes converging much faster than basic DA via parameter expansion over latent scales.
- PX-EM (Liu-Rubin-Wu 1998): Parameter expansion to accelerate EM: expand the complete-data model with unidentified working parameters, compute their complete-data MLE via standard formulas, then collapse back; achieves faster convergence than basic EM for t-distributions, probit, and robit models.
- CA-DA (Liu 1999): Covariance-adjusted data augmentation for MCMC — adjust the current draw by conditionally redrawing sufficient statistics of expanded parameters to integrate out nuisance scales; reduces chain autocorrelation and underpins E-DA 2 for robit.
Related