Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

Recurrent Neural Network (RNN) - Forget Layer, and TensorFlow

I'm new to RNN, and I'm trying to figure out the specifics of LSTM cells and they're relation to TensorFlow: Colah GitHub enter image description here Does the GitHub website's example uses the same LSTM cell compared to TensorFlow? The only thing I got on the TensorFlow site was that basic LSTM cells uses the following architecture: Paper If it's the same architecture then I can hand compute the numbers for a LSTM cell and see if it matches.

Also when we set a basic LSTM cell in tensorflow, it takes in a num_units according to: TensorFlow documentation

tf.nn.rnn_cell.GRUCell.__init__(num_units, input_size=None, activation=tanh)

Is this number of hidden state (h_t)) and cell state (C_t)?

According to the GitHub website, there isn't any mention the number of cell state and hidden states. I'm assuming they have to be the same number?

like image 577
user1157751 Avatar asked May 24 '17 10:05

user1157751


People also ask

What are two limitations of recurrent neural network?

Disadvantages Of RNNThe computation of this neural network is slow. Training can be difficult. If you are using the activation functions, then it becomes very tedious to process long sequences. It faces issues like Exploding or Gradient Vanishing.

How do RNNs remember?

RNN have a “memory” which remembers all information about what has been calculated. It uses the same parameters for each input as it performs the same task on all the inputs or hidden layers to produce the output. This reduces the complexity of parameters, unlike other neural networks.

What is the main problem of RNN?

RNNs suffer from the problem of vanishing gradients. The gradients carry information used in the RNN, and when the gradient becomes too small, the parameter updates become insignificant. This makes the learning of long data sequences difficult.

Can RNN have multiple hidden layers?

Definitely you can have multiple hidden layers in RNN.


1 Answers

Implementation looks the same as GRUCell class doc also points the same paper (specifically for gated) with link given in Colah's article. Parameter num_units is the number of cells (assuming that is the hidden layer) corresponds to output_size due property definition.

like image 157
hurturk Avatar answered Oct 06 '22 00:10

hurturk