Deep Dive into Math Behind Deep Networks

[headalgorithm]

https://towardsdatascience.com/https-medium-com-piotr-skalski92-deep-dive-into-deep-networks-math-17660bc376ba

[m0zg]

That's not a "deep dive" that's the disappointingly barest minimum.

[nafizh]

I would like to see someone explain the math behind recurrent neural networks. Feed-forward neural networks are fairly straight-forward, and there are many, many blog posts explaining them already.

[jing → nafizh]

I think this resource could be helpful:

http://colah.github.io/posts/2015-08-Understanding-LSTMs/

Essentially, RNNs and feed forward networks are very similar - RNNs are just "unrolled through time" and every timestep shares the same weights. The activations are slightly different as well, but the core concept is the same as feed forward networks; it's not a completely different concept or idea.

[pmalynin → nafizh]

If you unroll an RNN (which is what is done) then you have many copies of a single Feed Forward Network. Nothing fancy. The gradient gets accumulated.

[platz]

Backpropagation section was just a list of formulas, and "it's because of the chain rule".

[mjfl]

isn't it just gradient descent?

[ousta → mjfl]

yes just like a big chunk of what ML is.