CNNs share weights in convolutional layers. This means that the same filter is used for each receptive field in a layer and that these replicated units share the same parameterization (weight vector and bias) and form a feature map.
The following diagram shows three hidden units of a network belonging to the same feature map:
Figure 5.3: Hidden units
The weights in the darker gray color in the preceding diagram are shared and identical. This replication allows features detection regardless of the position they have in the visual field. Another outcome of this weight sharing is the following: the efficiency of the learning process increases by drastically reducing the number of free parameters to be learned.
