In RNN, Yt is the output which is given by the equation;
Yt = Why.ht
where Why is the weight associated with the output layer and ht is the current state.
What is the need of having this weight at the output layer of an RNN? Is it anything related to the memory of the network?
https://stats.stackexchange.com/questions/221513/why-are-the-weights-of-rnn-lstm-networks-shared-across-time
https://stats.stackexchange.com/questions/221513/why-are-the-weights-of-rnn-lstm-networks-shared-across-time
It is an important part of critical values, within all of the model performance located in weights. For understanding its importance, I suggest you to follow https://towardsdatascience.com/recurrent-neural-networks-d4642c9bc7ce
There are three basic components in a basic Neural Network (NN). The inputs, the hidden state, and the output. You can think of the hidden states as 'features' computed from the inputs. Then, you use those features to produce your prediction (output). You need a set of weights that map the inputs to the hidden state, and another set of weights that map the hidden representation to the output.
You can think of a Recurrent Neural Network as a series of basic neural networks. One per each time-step. The weights that map the input to the hidden state are going to be the same per every time point. The same is true for the weights that map the hidden state to the output. the only element that you have to add is another set of weights that connect the hidden state at time t, with the hidden states at time {t-1}. The weights that you refer to in your question are the ones that map the hidden state to the output. You need this weight to make the actual prediction.
- Badges
- Science topic