Thus, for the deepest network, we can construct an architecture consisting of three convolutional layers followed by a max pooling layer, three more convolutional layers, a second max pooling layer, then the fully connected layers.
It contains two convolutional connections followed by two pooling connections and three full connections.
Simonyan and Zisserman [31] proposed very deep neural networks which consist of more convolutional and pooling layers.
For the convolutional connections, we set the kernel size (m × n) to 3 × 1, the number of kernels (k) to 100, and the stride (s) to 1. Here, we consider the first convolutional layer as an example: the size of the input data is 99×× 1, and the first convolutional layer convolves these data with 100 3 × 1 kernels, with a stride (step size) of 1.
As HVS tends to select specified region of the visual field depending on saliency to interpret complex scenes, we use saliency data as an element-by-element mask on feature maps learned from convolutional connections.
Each convolutional connection is followed by a pooling layer that performs subsampling in the form of local maximization or averaging; such subsampling can reduce the dimensionality of the feature map and, furthermore, the sensitivity of the output.
This indicates that nodes with more connections are also more likely to have stronger connections.
More missed connections, more friction, more heat, more anger.
These nodes are those with more outgoing connections than incoming connections.
Public sharing allows for more serendipitous connections.
