In the 1960s, it was discovered that the visual cortex in animals doesn't act in a way deep feedforward networks do with images. Rather, a single neuron in the visual cortex is connected to a small region (and not a single pixel), which is called a receptive field. Any activity in the receptive field triggers the corresponding neuron.Â
Inspired by the receptive field in the visual cortex, scientists came up with the idea of local connectivity to reduce the number of artificial neurons required to process images. This modified version of deep feedforward networks was termed CNN (all through this book, CNN refers to convolutional neural network). In 1989, Yann LeCun developed a trainable CNN that was able to recognize handwritten digits. In 1998, again, Yann LeCun's LeNet-5 model successfully used seven stacked layers of convolution (like layers...
