Some more stuff for machine learning

[currymj]

I think there are some additional promising libraries in the area of machine learning.

• Most exciting from a propaganda perspective, there are Haskell bindings for Tensorflow. The library is hosted by the tensorflow organization but states that it is unofficial. It seems like it might be someone's 20% project.
• The ad library for automatic differentiation is arguably best-in-class. Computing derivatives/gradients is a fundamental operation for a lot of machine learning.
• There are several still-immature pure Haskell neural network libraries that use ad for backprop, including neural and grenade. The type system is used to ensure that input/output types of each layer of the network are consistent.
• The monad-bayes library is under rapid development (new commits ~daily) and has the potential to be very good for probabilistic programming. It is only on github now, won't be published to hackage until it is more stable.

I think ad could probably be considered mature. The rest less so.

[Gabriella439]

I added tensorflow, ad and grenade to the machine learning section in 3ad4a72

I left out neural and monad-bayes for now since they seemed to be lower quality than grenade, but I'll keep this issue open to remind myself to revisit them later on