2604.01408 Reparameterization of Non-Centered Hierarchical Models via Automatic Selection Improves NUTS Convergence by 4x: A Study Across 300 Posteriors
Non-centered parameterizations (NCPs) are widely recommended for hierarchical Bayesian models when group-level variance is small, yet the choice between centered and non-centered forms is typically manual. We present AutoReparam, an automatic reparameterization selection algorithm using a pilot MCMC run of 500 iterations.