【LSTM】Understanding LSTM

Original text: https://colah.github.io/posts/2015-08-Understanding-LSTMs/

recurrent neural network

Humans don;t start their thinking from scratch every second.
Humans don;t start their thinking from scratch every second.

• from scratch from scratch

As you read this article, your understanding of each word will be based on your understanding of the preceding words. You can't throw everything away and start thinking from scratch, your thoughts are continuous.

Traditional neural networks cannot do this, which seems to be a major drawback. For example, imagine that you want to classify the categories of events at each time point in a movie. Traditional neural networks may not be able to use the previous time to reason about the later time.

Recurrent neural networks solve this problem. Their network has a loop that allows information to persist.

In the image above, a neural network, A, has an input $x_t$and output a value $h_t$. Loops allow information to be passed from one step to the next.