On: The State of Machine Learning Frameworks

[phillipcarter]

What I really liked about this article is near the end, where it identifies two things:

* Automatic differentiation of higher-order differentiation being important, and how there's clearly room to disrupt there

* Increasing hardware diversity seems to mean that both frameworks will run into a brick wall as-is

Exciting space. It'll be fascinating to see how dramatically, or not, things change in the coming years.