“Huilian-cslt-week-2016-1-14”版本间的差异
来自cslt Wiki
（以“==Environment== ==RNN Poem Processing== * attention training from Renmin newspaper done --Qixin Wang done * change corpus done * contact specialists of Song Pe...”为内容创建页面） |
|||
| 第7行： | 第7行： | ||
* contact specialists of Song Peams done | * contact specialists of Song Peams done | ||
* write a patent done | * write a patent done | ||
| − | * submit a patent | + | * submit a patent done |
* demo release(unified model) done | * demo release(unified model) done | ||
* add to language model information | * add to language model information | ||
| − | * | + | * contact specialists |
| + | * train and check | ||
==RNN Writing Processing== | ==RNN Writing Processing== | ||
| 第18行： | 第19行： | ||
:* train using chinese dataset done | :* train using chinese dataset done | ||
* train reference to the paper of BAYESIAN PROBABILISTIC PROGRAM FOR IMAGE GENERATION | * train reference to the paper of BAYESIAN PROBABILISTIC PROGRAM FOR IMAGE GENERATION | ||
| + | * source code realsed | ||
==RNN QA== | ==RNN QA== | ||
2016年1月14日 (四) 01:56的版本
Environment
RNN Poem Processing
- attention training from Renmin newspaper done --Qixin Wang done
- change corpus done
- contact specialists of Song Peams done
- write a patent done
- submit a patent done
- demo release(unified model) done
- add to language model information
- contact specialists
- train and check
RNN Writing Processing
- write code done -- Chao Xing
- generate Xing style using variational RNN
- use mnist dataset done
- train using chinese dataset done
- train reference to the paper of BAYESIAN PROBABILISTIC PROGRAM FOR IMAGE GENERATION
- source code realsed
RNN QA
- rnn rescore
- rnn rescore should be used as a factor for learning to rank
- similar question --Bingdong
- Huilan test -- Bingdong
- train corpus of word vector--hulian
- add traditional method of learning to rank--Tianyi
- Huilan test -- Bingdong
- memory machine(about three months)--Tianyi
Couplet
- attention training done --Tianyi Luo
- focused attending training
- revise training corpus
