Convolutional Neural Networks 

One of the earliest applications of neural networks was demonstrated with Optical Character Recognition (OCR), but they were limited by time, computational resources, and other challenges faced when training bigger networks.

CNN is a part of feedforward neural networks, which are influenced by biological processes. This works in the same way that neurons work in the brain, as well as the connectivity patterns between them. These neurons will respond to stimuli that are only in a specific region in the visual field, known as the receptive field. When multiple neurons overlap each other, they will cover the whole visual field. The following diagram shows the CNN architecture:

CNN has an input layer and one output layer, as well as multiple hidden layers. These hidden layers consist of pooling layers, convolutional layers, normalization layers, and fully connected layers. Convolutional layers apply a convolution operation and pass the result to the next layer. This resembles how a neuron responds to its visual stimuli. Each neuron will reply to its receptive field only. Deep CNNs are used in various applications, including facial key-point detection, action classification, speech recognition, and so on.