Comment by mindcrime

The most obvious thing is understanding back-propagation. Backprop is pretty much all partial derivatives / chain rule manipulations. Also a lot of machine learning involves convex optimization which entails some calculus.