“Zhiyuan Tang 2015-03-30”版本间的差异
来自cslt Wiki
| （相同用户的一个中间修订版本未显示） | |||
| 第3行： | 第3行： | ||
preliminarily confirmed DNN's sparsity from the weights by Matlab and showed the weights' distribution graphicly; | preliminarily confirmed DNN's sparsity from the weights by Matlab and showed the weights' distribution graphicly; | ||
| + | |||
more systematic knowlege about ASR has been gained; | more systematic knowlege about ASR has been gained; | ||
| + | |||
CONTEXT was not yet studied carefully, :( | CONTEXT was not yet studied carefully, :( | ||
| 第10行： | 第12行： | ||
discover more things in the weights if it has more; | discover more things in the weights if it has more; | ||
| − | do some | + | |
| + | do some literature analysis about CONTEXT; | ||
| + | |||
go through the OBD-based sparse DNN-HMM. | go through the OBD-based sparse DNN-HMM. | ||
2015年3月30日 (一) 14:34的最后版本
Last week:
preliminarily confirmed DNN's sparsity from the weights by Matlab and showed the weights' distribution graphicly;
more systematic knowlege about ASR has been gained;
CONTEXT was not yet studied carefully, :(
This week:
discover more things in the weights if it has more;
do some literature analysis about CONTEXT;
go through the OBD-based sparse DNN-HMM.
