Three-part series about differentiable programming applied to machine learning in a purely functional typed setting, and the benefits of this combination.