Question

What are the dissimilarities between the properties of a fully connected network and that of a convolutional network

Answer 1

Fully connected network -

• this network takes input as a flattened single column vector.
• in such kind of network, each node in a layer is connected with every other node in the next layer.
• the total number of parameters increases to very high number
• such networks are insufficient as there is much redundancy in such networks
• spatial information is ignored in such network.
• used less nowadays as it is considered to be inefficient.

Convolutional Network -

• the weights used in this network are smaller as compared to fully connected networks
• the weights are shared in this network and hence small in size so used for higher dimensional inputs like image.
• it is much more effective than fully connected networks.
• the nodes among the layers are partially connected
• it can take both matrices and vectors as input (linient than fully connected networks)
• widely used nowadays