“Zhiyuan Tang 2015-09-07”版本间的差异
| 第8行： | 第8行： | ||
3. with Xuewei, reproduced the process of reorganizing a picture with the style of another specific picture | 3. with Xuewei, reproduced the process of reorganizing a picture with the style of another specific picture | ||
| − | on a laptop, while subject to | + | on a laptop, while subject to the laptop's memory, limited to inception net with sgd optimizer |
| + | |||
| + | (VGG network with the default L-BFGS optimizer consumes very much memory, which is better), | ||
then we moved the system to the grid, still the problem, while we want to solve it via software rather than hardware. | then we moved the system to the grid, still the problem, while we want to solve it via software rather than hardware. | ||
2015年9月7日 (一) 21:16的最后版本
Last week:
1. got the result of re-tuned bidirectional net on WSJ with dark knowledge, which is the best;
2. got more results of end-to-end ASR with B-LSTM on bigger data (1000+ hours) for basic comparison;
3. with Xuewei, reproduced the process of reorganizing a picture with the style of another specific picture
on a laptop, while subject to the laptop's memory, limited to inception net with sgd optimizer
(VGG network with the default L-BFGS optimizer consumes very much memory, which is better),
then we moved the system to the grid, still the problem, while we want to solve it via software rather than hardware.
This week:
1. rebuild the neural art style transfer system on the grid totally, and do some further modification;
2. other regular experiments.
