“Qixin Wang 2016-01-04”版本间的差异
来自cslt Wiki
| （相同用户的2个中间修订版本未显示） | |||
| 第5行： | 第5行： | ||
* Attention problem in: | * Attention problem in: | ||
** Hidden reverse | ** Hidden reverse | ||
| + | *** b = a[::-1] | ||
| + | *** b = a.dot(diag(1).T) | ||
[[文件:Snip20160104 1.png]] | [[文件:Snip20160104 1.png]] | ||
** start end signal | ** start end signal | ||
| 第15行： | 第17行： | ||
pre trained: | pre trained: | ||
[[文件:Snip20160103 1.png|1000px]] | [[文件:Snip20160103 1.png|1000px]] | ||
| + | |||
| + | mem: | ||
| + | [[文件:Snip20160104 2.png|1000px]] | ||
=== Plan to do next week === | === Plan to do next week === | ||
* Finish other experiments | * Finish other experiments | ||
* Compare experiments | * Compare experiments | ||
2016年1月4日 (一) 01:19的最后版本
Work done in this week
- Mikolov Language Model merge experiment in:
- Character level, sentence level, merge way
- Attention problem in:
- Hidden reverse
- b = a[::-1]
- b = a.dot(diag(1).T)
- Hidden reverse
- start end signal
- Go information
- Modern vector
pure:
pre trained:
mem:
Plan to do next week
- Finish other experiments
- Compare experiments
