Computer scienceData scienceMachine learningIntroduction to deep learningRecurrent neural networks

LSTM

Activation functions in LSTM gates and states

Report a typo

Why are sigmoid functions used in LSTM gates, while the hyperbolic tangent ( $\tanh$ ) function is used for the hidden state $H_t$ and input node?

Select one option from the list

Sigmoid functions output a binary decision for gates, while tanh provides a full range of activations for states

Sigmoid functions are used to normalize the input data, while tanh accelerates the learning process for the states

Tanh functions are computationally simpler for gates, while sigmoids provide more computational stability for states

Sigmoid functions restrict the gate outputs to a range of 0 to 1 for controlled gating, while tanh normalizes the states' outputs between -1 and 1

___

Create a free account to access the full topic