本文介绍了如何使用循环神经网络(RNN)处理不同类型的序列数据，包括一对一、一对多、多对一及多对多的场景，并详细探讨了RNN的工作原理、截断回传误差法以及长短期记忆网络(LSTM)解决梯度消失问题的方法。
Recurrent Neural Networks: Process Sequences
- one to one: Vanilla Neural Networks. raw
- one to many: eg. Image Captioning image -> sequence of words
- many to one: e.g. Sentiment Classification sequence of words -> sentiment
- many to many: e.g. Machine Translation seq of words -> seq of words
(Vanilla) Recurrent Neural Network
ht=f(ht−1,xt)h_t = f(h_{t-1},x_t)ht=f(ht−1,xt)
ht=tanh(Whht−1+Wxxt)h_t = tanh(W_{h} h_{t-1}+W_{x}x_t)ht=tanh(Whht−1+Wxxt)
y=Wyhty = W_yh_ty=Wyht
Truncated backpropagation through time
- Run forward and backward through chunks of the sequence instead of whole sequence
Image Captioning with Attention
- CNN网络生成L个D维的feature，代表L个location的feature
- RNN迭代每一步生成一个L个位置的分布向量，表示图片中L个位置中每个位置的attention权重
Long Short Term Memory (LSTM)
Difference to avoid gradient vanishment:
- Backpropagation from ctc_tct to ct−1c_{t-1}ct−1 only elementwise multiplication by f, no matrix multiply by W.
- f is different at every step. In Vanilla RNN, it always multiply the same matrix.
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
