与nlp相关的模型概览

末蓝、 2023-05-29 02:44 2阅读 0赞

### 与nlp相关的模型概览 ###

*  word2vec
 *   *  CBOW
     *  Skip-Gram
 *  Doc2Vec
 *  RNN、LSTM、GRU
 *   *  循环神经网络RNN
     *  长短期记忆神经网络（Long Short Term Memory，LSTM）
     *  门控循环单元（Gated Recurrent Unit，GRU）
 *  机器翻译
 *   *  Encoder-Decoder
     *  Attention-based Encoder-Decoder
     *   *  Global Attention
         *  Local Attention
         *  Self Attention
         *  Hierarchical Attention
     *  Bi-Directional Encode layer
     *  Residual Encode layer
     *  总结
 *  Image Caption
 *  Transformer
 *  Bert(Bidirectional Encoder Representations from Transformers)
 *  分词
 *   *  基于BiLSTM+CRF的分词系统
 *  文本分类
 *   *  简单模型
     *  LSTM
     *  双向RNN+残差网络
     *  堆叠循环神经网络
     *  CNN
     *  lstm-attention

# word2vec #

## CBOW ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 1]

## Skip-Gram ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 2]

# Doc2Vec #

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 3]

# RNN、LSTM、GRU #

## 循环神经网络RNN ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 4]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 5]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 6]

## 长短期记忆神经网络（Long Short Term Memory，LSTM） ##

计算量大  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 7]

## 门控循环单元（Gated Recurrent Unit，GRU） ##

相比LSTM, 计算量小一些  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 8]

# 机器翻译 #

## Encoder-Decoder ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 9]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 10]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 11]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 12]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 13]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 14]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 15]

## Attention-based Encoder-Decoder ##

### Global Attention ###

encoder的全部信息  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 16]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 17]  
![\*\*v\*\*][v]

### Local Attention ###

encoder的局部信息  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 18]

### Self Attention ###

传统的attention只有source（encoder）和target（decoder）之间的关联关系，忽略了source和target端分别的关联关系，Self Attention可以捕捉到词与词之间的关系。  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 19]

### Hierarchical Attention ###

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 20]

## Bi-Directional Encode layer ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 21]

## Residual Encode layer ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 22]

## 总结 ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 23]

# Image Caption #

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 24]

# Transformer #

[http://nlp.seas.harvard.edu/2018/04/03/attention.html][http_nlp.seas.harvard.edu_2018_04_03_attention.html]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 25]

# Bert(Bidirectional Encoder Representations from Transformers) #

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 26]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 27]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 28]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 29]

# 分词 #

## 基于BiLSTM+CRF的分词系统 ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 30]

# 文本分类 #

## 简单模型 ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 31]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 32]

## LSTM ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 33]

## 双向RNN+残差网络 ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 34]

## 堆叠循环神经网络 ##

堆叠的目的：  
底层rnn捕捉的是基本的信息，如：词的含义或者词性  
高层rnn捕捉的是更深层次的信息，如：语义  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 35]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 36]

## CNN ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 37]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 38]

## lstm-attention ##

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 39]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20191101171618682.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20191101171806294.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20191101172216637.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20191102215829669.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20191101172815120.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20191101173018367.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/20191101181022842.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 7]: https://img-blog.csdnimg.cn/20191101181246654.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 8]: https://img-blog.csdnimg.cn/20191101181342696.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 9]: https://img-blog.csdnimg.cn/20191101181451938.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 10]: https://img-blog.csdnimg.cn/20191101181547189.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 11]: https://img-blog.csdnimg.cn/20191101181650434.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 12]: https://img-blog.csdnimg.cn/20191101181725736.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 13]: https://img-blog.csdnimg.cn/20191101181748666.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 14]: https://img-blog.csdnimg.cn/20191101181950355.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 15]: https://img-blog.csdnimg.cn/20191101182005799.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 16]: https://img-blog.csdnimg.cn/20191101182119681.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 17]: https://img-blog.csdnimg.cn/20191101182514678.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[v]: https://img-blog.csdnimg.cn/20191104211043565.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 18]: https://img-blog.csdnimg.cn/20191101183956816.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 19]: https://img-blog.csdnimg.cn/20191101203330294.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 20]: https://img-blog.csdnimg.cn/20191101205117912.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 21]: https://img-blog.csdnimg.cn/20191101184251924.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 22]: https://img-blog.csdnimg.cn/20191101184513906.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 23]: https://img-blog.csdnimg.cn/20191101184624254.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 24]: https://img-blog.csdnimg.cn/20191101182349184.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[http_nlp.seas.harvard.edu_2018_04_03_attention.html]: http://nlp.seas.harvard.edu/2018/04/03/attention.html
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 25]: https://img-blog.csdnimg.cn/2019110122462065.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 26]: https://img-blog.csdnimg.cn/20191102182022925.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 27]: https://img-blog.csdnimg.cn/20191102182041125.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 28]: https://img-blog.csdnimg.cn/20191102182202301.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 29]: https://img-blog.csdnimg.cn/20191102182130424.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 30]: https://img-blog.csdnimg.cn/20191102205959410.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 31]: https://img-blog.csdnimg.cn/20191102224301735.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 32]: https://img-blog.csdnimg.cn/20191102223313680.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 33]: https://img-blog.csdnimg.cn/20191102225841761.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 34]: https://img-blog.csdnimg.cn/20191102230228317.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 35]: https://img-blog.csdnimg.cn/20191102230554808.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 36]: https://img-blog.csdnimg.cn/20191102230804909.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 37]: https://img-blog.csdnimg.cn/201911041512060.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 38]: https://img-blog.csdnimg.cn/20191104152139979.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ_size_16_color_FFFFFF_t_70 39]: https://img-blog.csdnimg.cn/20191104205612609.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0hlYXJ0X1NlYQ==,size_16,color_FFFFFF,t_70