GVSSB a parameter-expanded Coordinate-Ascent Variational Inference algorithm for high-dimensional linear regression models with grouped variables under spike-and-slab priors in R language. The sparse linear models can be formulated as

$$\bm{Y} = \sum_{j = 1}^G\bm{X}_j\bm{\beta}+\bm{\epsilon},\quad \bm{\epsilon}\sim N(\bm{0}, \sigma^2 \bm{I}_n)$$

where the coefficient $\bm{\beta}$ is sparse with most of its elements being zero. It is also capable of solving the sparse additive model

$$Y_i=b+\sum_{j=1}^pf_j(X_{ij})+\epsilon_i,\quad \epsilon_i\sim N(0,\sigma^2)$$

with most of $f_j$’s being zero functions by spline fitting. The smothness of splines are regularized to prevent overfitting. For more information please refer here.