Option A holds as the goal of regularization is to reduce model complexity and thereby mitigate overfitting. To see that Option B holds, we first note that the $l_o$-norm is in fact neither a true norm nor a convex function. Moreover, by virtue of the triangle inequality any $l_p$-norm with $p \geq 1$ is convex. The $l_1$-norm is the closest convex approximation to the $l_o$-norm. Option C is false because it is not possible to write a minimizer of Lasso in closed form, though its solution can be expressed as the solution to a quadratic program. Option D holds as Ridge regression tends to penalize larger components of the weight vector more than smaller values. This is a consequence of the square-operator inherent to ridge.
