Lasso Meets Horseshoe
The goal of our paper is to survey and contrast the major advances in two of the most commonly used high-dimensional techniques, namely, the Lasso and horseshoe regularization methodologies. Lasso is a gold standard for best subset selection of predictors while the horseshoe is a state-of-the-art Bayesian estimator for sparse signals. Lasso is scalable and fast using convex optimization whilst the horseshoe is a non-convex penalty. Our novel perspective focuses on three aspects, (i) efficiency and scalability of computation and (ii) methodological development and performance and (iii) theoretical optimality in high dimensional inference for the Gaussian sparse model and beyond.
PDF Abstract