network in network---NIN&mlpConv

mlpConv:

add some 1*1 conv(ReLU) after the conv, the n can be seen as the number of  1*1 conv(ReLU)  in mlpConv

------------------------------

global average pooling  without fc layer

the number of feature maps before loss-function(eg. softmaxwithloss) must equal to the category's ( in the classification task). and each feature map after ave pooling feed to the loss func. 

----------------------Reference--------------------------

[1] Min Lin, Qiang Chen, Shuicheng Yan.  Network In Network