Variable Selection and Parameter Estimation Using a Continuous and Differentiable Approximation to the L0 Penalty Function