https://blog.youkuaiyun.com/lebulazh-cnhttps://blog.youkuaiyun.com/https://blog.youkuaiyun.com/lebula/article/details/78898119https://blog.youkuaiyun.com/lebula/article/details/78898119lebulaTue, 26 Dec 2017 00:17:02 +0800https://blog.youkuaiyun.com/lebula/article/details/70738309https://blog.youkuaiyun.com/lebula/article/details/70738309lebulaTue, 25 Apr 2017 13:29:09 +0800https://blog.youkuaiyun.com/lebula/article/details/53661595https://blog.youkuaiyun.com/lebula/article/details/53661595lebulaThu, 15 Dec 2016 04:04:56 +0800t,通常作用域这些序列组成的minibatch 2.如果要求h(t)足够丰富，可以近似恢复输入序列，则相当于auto encoder. 3.三种重要的RNN设计模式： 1）每个时间步都有输出，且在隐单元间有循环链接 2）每个时间步都有输出，且在当前时序的输出到下一时刻的隐单元之间有循环链接 3）隐单元间有循环连接，全部序列只有一个输出]]>https://blog.youkuaiyun.com/lebula/article/details/53419383https://blog.youkuaiyun.com/lebula/article/details/53419383lebulaThu, 01 Dec 2016 11:06:23 +0800https://blog.youkuaiyun.com/lebula/article/details/53401947https://blog.youkuaiyun.com/lebula/article/details/53401947lebulaTue, 29 Nov 2016 23:18:08 +0800https://blog.youkuaiyun.com/lebula/article/details/53384936https://blog.youkuaiyun.com/lebula/article/details/53384936lebulaMon, 28 Nov 2016 21:17:56 +0800https://blog.youkuaiyun.com/lebula/article/details/53172817https://blog.youkuaiyun.com/lebula/article/details/53172817lebulaTue, 15 Nov 2016 16:58:01 +0800https://blog.youkuaiyun.com/lebula/article/details/53168712https://blog.youkuaiyun.com/lebula/article/details/53168712lebulaTue, 15 Nov 2016 14:19:02 +0800 {member1, | +-------+ | | | | input +----+----> member2, | +-------+ | | |]]>https://blog.youkuaiyun.com/lebula/article/details/53118543https://blog.youkuaiyun.com/lebula/article/details/53118543lebulaThu, 10 Nov 2016 18:08:04 +0800https://blog.youkuaiyun.com/lebula/article/details/52955076https://blog.youkuaiyun.com/lebula/article/details/52955076lebulaFri, 28 Oct 2016 12:04:32 +0800https://blog.youkuaiyun.com/lebula/article/details/52928843https://blog.youkuaiyun.com/lebula/article/details/52928843lebulaTue, 25 Oct 2016 23:23:47 +0800https://blog.youkuaiyun.com/lebula/article/details/52883608https://blog.youkuaiyun.com/lebula/article/details/52883608lebulaFri, 21 Oct 2016 15:19:31 +0800https://blog.youkuaiyun.com/lebula/article/details/52865285https://blog.youkuaiyun.com/lebula/article/details/52865285lebulaThu, 20 Oct 2016 00:18:45 +0800https://blog.youkuaiyun.com/lebula/article/details/52856398https://blog.youkuaiyun.com/lebula/article/details/52856398lebulaWed, 19 Oct 2016 09:27:56 +0800https://blog.youkuaiyun.com/lebula/article/details/52854574https://blog.youkuaiyun.com/lebula/article/details/52854574lebulaWed, 19 Oct 2016 00:12:04 +0800https://blog.youkuaiyun.com/lebula/article/details/52854247https://blog.youkuaiyun.com/lebula/article/details/52854247lebulaTue, 18 Oct 2016 23:11:24 +0800https://blog.youkuaiyun.com/lebula/article/details/52847353https://blog.youkuaiyun.com/lebula/article/details/52847353lebulaTue, 18 Oct 2016 11:27:20 +0800https://blog.youkuaiyun.com/lebula/article/details/52847040https://blog.youkuaiyun.com/lebula/article/details/52847040lebulaTue, 18 Oct 2016 11:02:35 +0800 use MLP (can approximate any functions, whereas maxout only ]]>https://blog.youkuaiyun.com/lebula/article/details/52844739https://blog.youkuaiyun.com/lebula/article/details/52844739lebulaTue, 18 Oct 2016 01:30:13 +0800https://blog.youkuaiyun.com/lebula/article/details/52832190https://blog.youkuaiyun.com/lebula/article/details/52832190lebulaSun, 16 Oct 2016 19:01:32 +0800