Although regular hidden layers (we also call them fully connected layers) do the job of obtaining representations at certain levels, these representations might be able to help us differentiate between images of different classes. We need to extract richer and distinguishable representations that, for example, make a "9" a "9", a "4" a "4", or a cat a cat, a dog a dog. We resort to CNNs as variants of multi-layered neural networks which are biologically inspired by the human visual cortex. Basically, CNNs take inspiration from the following two neuroscience findings:
- The visual cortex has a complex system of neuronal cells that are sensitive to specific sub-regions of the visual field, called the ...
