Deep Learning and a New Programming Paradigm – Towards Data Science

I like where this line of thoughts goes: functional programming means functional composability. Decoupled deep learning modules is an exciting research area: Decoupled Neural Interfaces using Synthetic Gradients has shown, for example, very promising results [22] I am not sure if the term Differentiable Programming will stick around. The risk of confusion with Differential Dynamic Programming is high. The idea, on the other hand, is intriguing. Very intriguing and I am very happy to see projects such as Tensorlang [17] gaining traction. Wired argued that soon we won't program computers. We'll train them like dogs.