October 2017
Beginner to intermediate
270 pages
7h
English
The origin of RNNs is surprisingly common to the other modern neural network architectures, dating back to Hopfield networks from the 1980s, but with counterparts in the 1970s.
The common structure for the first iterations of the recurrent architecture can be represented in the following way:

Classic RNN nodes have recurrent connections to themselves, and so they can evolve their weights as the input sequence progresses. Additionally, on the right of the diagram, you can see how the network can be unrolled to generate a set of outputs based on the stochastic model it has saved internally. It stores ...
Read now
Unlock full access