Bayesian Effective Dimension from Information Growth

Sayantan Banerjee

Nominal dimension can greatly overstate how much a Bayesian model actually learns from data. We quantify this gap through a prior-dependent effective dimension based on parameter--…