January 2018
Beginner to intermediate
284 pages
8h 35m
English
The next step is to update the cell state from
to
. The selected hidden memory from step 1 is combined additively with a filtered version of the current input. The filtering is conducted again by the so-called input gate, which is a sigmoid layer that decides which values we’d like to update. This filtering decision is multiplied by the activation results by tanh and then added to the selected memory vector from the forget gate. The result is used to update the cell state .
Read now
Unlock full access