Animated RNN, LSTM, and GRU Computation Process

Animated RNN, LSTM, and GRU Computation Process

by
Source | Zhihu Author | JerryFly
Link | https://zhuanlan.zhihu.com/p/115823190
Editor | Deep Learning Matters WeChat Official Account
This article is for academic exchange only. If there is any infringement, please contact us for deletion.
RNN is commonly used to handle sequential problems. This article demonstrates the computation process of RNN using animated graphics. The three most common RNN architectures are:

  • vanilla RNN

  • long short-term memory (LSTM)

  • gated recurrent units (GRU)

Below are the descriptions of the various legends used in the illustrations.
Animated RNN, LSTM, and GRU Computation Process
NOTE: The animation shows the processing inside the RNN cell at a certain moment (indexed by t). In the example, input size=3, hidden units=2, batch size=1.

Vanilla RNN

Animated RNN, LSTM, and GRU Computation Process
Animated RNN, LSTM, and GRU Computation Process

  • t–time step

  • X–input

  • h–hidden state

  • length of X–dimension of input

  • length of h–number of hidden units

LSTM

Animated RNN, LSTM, and GRU Computation Process
Animated RNN, LSTM, and GRU Computation Process

  • C–cell state

GRU

Animated RNN, LSTM, and GRU Computation Process
Animated RNN, LSTM, and GRU Computation Process
Original link: https://towardsdatascience.com/animated-rnn-lstm-and-gru-ef124d06cf45

—End—

Recommended for you
MIT President Comments on Sino-US Tech Competition: Victory is not Hoping for Opponent's Defeat
GitHub Major Update: Online Development is Live, It's Time to Uninstall IDE
US Officially Announces 117,000 IT Workers Unemployed, Truly Unprecedented!

23 Powerful Pandas Codes Commonly Used in Data Analysis
Trump Targets H1B Visa, LeCun and Ng Among Others Publicly Condemn!

Leave a Comment