Réseau récurrent à portes


Domaine

Vocabulary Apprentissage profond

Définition

Termes privilégiés

Anglais

GRU

The Gated Recurrent Unit is a simplified version of an LSTM unit with fewer parameters. Just like an LSTM cell, it uses a gating mechanism to allow RNNs to efficiently learn long-range dependency by preventing the vanishing gradient problem. The GRU consists of a reset and update gate that determine which part of the old memory to keep vs. update with new values at the current time step. • Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation • Recurrent Neural Network Tutorial, Part 4 – Implementing a GRU/LSTM RNN with Python and Theano