A novel nonconvex, smooth-at-origin penalty for statistical learning