Learning Interpretable Differentiable Logic Networks