There are convolution layers, pooling layers, and possibly a classifier layer (e.g. softmax layer) in a convolutional neural network (CNN).
I heard that there is also a fully-connected layer. What is that?
$\begingroup$
$\endgroup$
asked Oct 10 '14 at 20:32
-
1$\begingroup$ stats.stackexchange.com/a/182122/103153 $\endgroup$ – Lerner Zhang Oct 10 '17 at 4:33
$\begingroup$
$\endgroup$
Every neuron from the previous layer is connected to every neuron on the next layer1.
$\begingroup$
$\endgroup$
The convolutional and the Pooling layers create a feature space, and the flatten FUlly connected layer can be thought as a cheap way of learning a linear function out of the feature space. For the convenience of understanding think of it as a PCA that selects the good features among the feature space created by the Conv and POOL layer.
