25 Jan, 2016

Early Shrinkage Priors

Global Local

  • computationally easy
  • generally no mixing issues (unlike spike and slab)

Bayesian Lasso prior (Park and Casella, 2008)

  • bayesian adaptation of frequntist lasso
  • choose $\lambda$ using empirical bayes
  • Inverse-Gaussian distribution is not a obtained by taking the reciprocal of a gaussian! Misnomer.