http://index.cslt.org/mediawiki/index.php?action=history&feed=atom&title=Mengyuan_Zhao_14-12-07Mengyuan Zhao 14-12-07 - 版本历史2026-04-07T14:46:37Z本wiki的该页面的版本历史MediaWiki 1.23.3http://index.cslt.org/mediawiki/index.php?title=Mengyuan_Zhao_14-12-07&diff=12850&oldid=prevZhaomy：以“Weekly report: *sinovoice #8k-1400h-dropout-AM, 1400h-maxout-AM, results on sinovoice-wiki. #8k-4000h-AM-no-delta, mpe1 done, mpe2 running. results on sinovoice-wiki...”为内容创建页面2014-12-08T01:41:05Z<p>以“Weekly report: *sinovoice #8k-1400h-dropout-AM, 1400h-maxout-AM, results on sinovoice-wiki. #8k-4000h-AM-no-delta, mpe1 done, mpe2 running. results on sinovoice-wiki...”为内容创建页面</p>
<p><b>新页面</b></p><div>Weekly report:<br />
*sinovoice<br />
#8k-1400h-dropout-AM, 1400h-maxout-AM, results on sinovoice-wiki.<br />
#8k-4000h-AM-no-delta, mpe1 done, mpe2 running. results on sinovoice-wiki.<br />
#8k-4000h-AM-add-delta, mpe1 still running.<br />
#add-1-hidden-layer training. do some experiments on 100h-uyghur data, performance seems ok, but speed isn't fast. training on 8k-4000h, now.<br />
*cslt<br />
#DAE: training nnet to eliminate noise (XinWenLianBo-music). do some experiments, performance is ok. intend to trial on "echo". <br />
#investigate the effect of dropout on different proportion of noise data on training set. get some results.</div>Zhaomy