Understanding Convolutional Neural Networks (CNN) in Deep Learning

Convolutional Neural Networks - What is CNN in Deep Learning is a part of the "Learn A Concept" series. CNN uses deep learning techniques that simulate the behavior of the Visual Cortex in a human or animal brain that helps remember and identify images, objects and complex data.
CNN is a special type of Neural Network in Deep Learning primarily used in complex problems like Image Analysis and Audio analysis.
As you may be aware, computers store and visualize images and audio data as just a bunch of numbers in a specific format.

How does a complex input data like image or audio get converted to numbers? For example, when dealing with a 128 by 128 pixel image, the program creates a matrix of the same size and stores the intensity of the pixels in each cell.
CNN Architecture has convolution layers which is nothing but filters, applied using Matrix multiplication
Since the amount of data becomes huge with multiplication, the architecture uses a layer called as max-pooling layer to reduce the calculations by removing unwanted data

So, by applying various layers of filters and max pooling, the program is able to extract specific features that are common to each set of sample data.
What do we mean by extracting features? Let us feed a lot of pictures of Brad Pitt to the program. All these pictures will all have some common features that the program will identify like the shape of his nose, distance between eyes or size of the forehead. Each of these attribute is a set of numbers and used for classification when a new picture of Brad Pitt is presented.

CNN, works by applying various filters to the input data. When you edit images on your phone or a computer, you would have used some filters like Edge Detection, image Blur and Sharpening which are actually numerical calculations applied on the image.
This process actually helps in identifying critical and repeating features in the data.