Which combination of gates and operations in LSTM ensures that long-term dependencies
are preserved better than in traditional RNNs?

Question

Which combination of gates and operations in LSTM ensures that long-term dependencies 
 are preserved better than in traditional RNNs?

Accepted Answer

c.  Sigmoid forget gate + tanh state scaling + multiplicative gating

Answer

b.  ReLU activation and dropout layer

Answer

a.  Softmax output and ReLU activation

Answer

d.  Linear transformation followed by batch normalization

Crowdly