“2019-01-02”版本间的差异

来自cslt Wiki
跳转至： 导航、 搜索
 
（5位用户的8个中间修订版本未显示）
第10行： 第10行：
 
|Yibo Liu
 
|Yibo Liu
 
||  
 
||  
−
*  
+
* Tried to reconstruct Song Iambics generation on the previous code.
 
||
 
||
−
*  
+
* Read about GNN and try to relate it to the poem model.
 +
* Run BERT and apply to the existing model.
 
||
 
||
 
*   
 
*   
第38行： 第39行：
 
|Jiayao Wu
 
|Jiayao Wu
 
||  
 
||  
−
*  
+
* read some papers about kws and write the chapter before this weekend.
 +
* carding the TDNN-F structure and count the lines of weights to prune and do value prune first.
 
||
 
||
−
*  
+
* finish the assigned chapter of Speech book
 +
* finish the value prune experiment
 
||  
 
||  
 
*  
 
*  
第51行： 第54行：
 
|Zhaodi Qi
 
|Zhaodi Qi
 
||  
 
||  
−
*  
+
* read some papers about DID and other
 +
* Run language recognition task
 
||
 
||
−
*  
+
* Continue to complete the language recognition task
 
||
 
||
 
*   
 
*   
第64行： 第68行：
 
|Jiawei Yu
 
|Jiawei Yu
 
||  
 
||  
−
*  
+
* do some Literature survey on spoken Emotion Recognition for speech book
 +
* done for extract d-vector and phone vector for max-margin
 
||
 
||
−
*  
+
* run max-margin recipe for extracted vector
 +
* Consider how to implement the network structure
 +
* finishi the speech book of spoken Emotion Recognition
 
||
 
||
 
*   
 
*   
第75行： 第82行：
 
|Yunqi Cai
 
|Yunqi Cai
 
||  
 
||  
−
*  
+
*Run through the Thchs30 data, think about the research plan.
 
||
 
||
−
*  
+
*Get to know every step of the ASR processes, and investigate the RNN LM.
 
||
 
||
 
*  
 
*  
第86行： 第93行：
 
|Dan He
 
|Dan He
 
||  
 
||  
−
*  
+
*Compared the size of a fully connected layer after tt decomposition through experiment, at the same time compared the train loss, validation loss and test accuracy between the two case.
 
||
 
||
−
*
+
*Further compared the inference speed, the precision after retrain.
 +
*Try to decompose more layers of fully connected layers.
 
||
 
||
 
*
 
*

2019年1月3日 (四) 04:50的最后版本

People Last Week This Week Task Tracking (DeadLine)
Yibo Liu
  • Tried to reconstruct Song Iambics generation on the previous code.
  • Read about GNN and try to relate it to the poem model.
  • Run BERT and apply to the existing model.
Xiuqi Jiang
  • Made further adjustments to code, deleting unnecessary files and uploading generated samples to dir 'predict/results'.
  • More models have been trained and compared.
  • Collated notes on ML book and uploaded them on wiki.
  • Try to train a model generating length-unfixed texts.
Jiayao Wu
  • read some papers about kws and write the chapter before this weekend.
  • carding the TDNN-F structure and count the lines of weights to prune and do value prune first.
  • finish the assigned chapter of Speech book
  • finish the value prune experiment
Zhaodi Qi
  • read some papers about DID and other
  • Run language recognition task
  • Continue to complete the language recognition task
Jiawei Yu
  • do some Literature survey on spoken Emotion Recognition for speech book
  • done for extract d-vector and phone vector for max-margin
  • run max-margin recipe for extracted vector
  • Consider how to implement the network structure
  • finishi the speech book of spoken Emotion Recognition
Yunqi Cai
  • Run through the Thchs30 data, think about the research plan.
  • Get to know every step of the ASR processes, and investigate the RNN LM.
Dan He
  • Compared the size of a fully connected layer after tt decomposition through experiment, at the same time compared the train loss, validation loss and test accuracy between the two case.
  • Further compared the inference speed, the precision after retrain.
  • Try to decompose more layers of fully connected layers.
Yang Zhang
  • Released and published our WeChat app(星云听).
  • cleaned my code
  • wrote document for this project (have not finished yet).
  • continue to write document.
取自“http://cslt.org/mediawiki/index.php?title=2019-01-02&oldid=31827”