aAT1t9p7ShM
note:
neral network for example: intput --> 1 layer --> 2 layer--> 3layer --> 4 layer --> output
layer 1:have many feature maps input into layer 2.
layer2: have feature maps ,each feature map connect with layer1, do many convolution, each convolution have the same parameter.
That is, parameter sharing is within a feature map.
each feature map in layer2 will detect the same feature in all layer1 feature maps, so the feature move , layer2 will have the same result..
lay1's feature maps becomes input chinnels to layer2.