RNN¶
tutorial:
RNN series (TODO)
普通RNN的问题¶
在实际实现中,RNN 无法解决“long-term dependencies.”
Long Short Term Memory(LSTM) 1¶
Remembering information for long periods of time is practically their default behavior, not something they struggle to learn!
LSTM的巧妙之处在于通过增加输入门限,遗忘门限和输出门限,使得自循环的权重是变化的,这样一来在模型参数固定的情况下,不同时刻的积分尺度可以动态改变,从而避免了梯度消失或者梯度膨胀的问题。
Motivation¶
long-term dependency problem
应用¶
speech recognition
language modeling
translation
image captioning
…
- 1
Hochreiter S, Schmidhuber J. Long short-term memory[J] . Neural computation, 1997, 9(8): 1735-1780.