December 2018
Intermediate to advanced
158 pages
3h 58m
English
For artificial neurons in feedforward networks, the flow of activation is simply from the input to the output. Recurrent artificial neurons (RANs) have a connection from the output of the activation layer to its linear input, essentially summing the output back into the input. A RAN can be unrolled in time: each subsequent state is a function of previous states. In this way, a RAN can be said to have a memory of its previous states:

In the preceding diagram, the diagram on the left illustrates a single recurrent neuron. It sums its input, x, with the output, y, to produce a new output. The diagram on the right ...
Read now
Unlock full access