有没有喜欢神经网络学习的同好~

wwtpcsuper 发表于 2012-10-4 09:42
有别的network可以做

版主大大求解~

hetong_007 发表于 2012-10-4 10:34
版主大大求解~

google network science

neural network只是其中很小一个部分

wwtpcsuper 发表于 2012-10-4 09:42
有别的network可以做

哈哈果然看到熊哥了~

Ber 发表于 2012-10-4 10:42
哈哈果然看到熊哥了~

哈哈哈哈

wwtpcsuper 发表于 2012-10-4 22:38
google network science

neural network只是其中很小一个部分

看了wiki感觉只要是graph的模型都能叫做network science

hetong_007 发表于 2012-10-4 14:48
看了wiki感觉只要是graph的模型都能叫做network science

我也不懂。大牛求带。。

too many research areas are hot or getting hot now..but neural network is kinda ideal machine..it has been hot like 80s 90s..now not easy to really dig some good ideas..but if LZ really like it, just go ahead..maybe you can push the whole area forward a lot someday :-)

wwtpcsuper 发表于 2012-10-5 04:40
我也不懂。大牛求带。。

_(:з」∠)_版主巨巨别这样

solomonking 发表于 2012-10-5 07:15
too many research areas are hot or getting hot now..but neural network is kinda ideal machine..it ha ...

I know recently Google trained a nerual network system with 16000 computers to identify cats = = maybe in the future "extreme network" could make a difference

hetong_007 发表于 2012-10-5 13:03
I know recently Google trained a nerual network system with 16000 computers to identify cats = = m ...

Neural net已经沉寂了很多年了，但是最近deep belief network开始热起来，并且因为Neural Net之前被卡住在一定程度上是当时运算能力有限造成的，现在加了sparsity等各种限制以后发现貌似可以计算了，于是被拿来当autoencoder用在deep belief network里面。

Google搞得那个训练猫的基本上就是这么一回事儿，不过他们动用了Google的计算资源训练一个（好像是）9层的Nerual net，训练了超过一个星期，几千核的cluster不是你想搞想搞就能搞的= =当然可以在数学角度继续优化，不过那个就有点太理论且跑偏了