Home / Applications / Convolutional Neural Network

Convolutional Neural Network

Jenna Phipps
Last Updated May 24, 2021 8:03 am

A convolutional neural network is a technological system in which a machine learns to recognize the contents of images for better data processing. Its name refers to its complexity (multiple convolutions, or layers, of an image which the machine must analyze) and its likeness to the brain’s neural network. Computer neural networks attempt to imitate the way that the human brain analyzes the imagery that the eyes see and are considered a form of deep learning.

How does a convolutional neural network work?

The receivers, or filters, that observe and analyze an image are also called neurons, named for their biological counterparts. They are responsible for one particular area of an image. The images are separated into sections by height, width, and depth (if the color scale is RGB, then the image will have at least three layers). Each layer has matrices of numbers; the numbers are based on the pixelation in that section of the image.

A convolutional neural network (or CNN) maps each section of a digital image and calculates the numbers in each matrix, which correspond to features in the image. In a CNN, the neurons receive input from the first layer (which is typically general information such as the edges and curves in an image) and move through the layers sequentially. The area which neurons cover on the image expands as the system moves through more layers.

The process of training machines to initially recognize features in images is called backpropagation. It requires going back through the system’s image analysis and figuring out what didn’t work.

How are neural networks used?

Neural networks are one of the most essential technologies in deep learning, a subset of machine learning that trains computers through visual exposure. The purpose is to make the machine more intelligent so that it can better process data. Facebook and Instagram are two examples of enterprises that use machine learning heavily for image processing. Facial recognition is one big use case, but other important uses fall into the category of business data analytics for sales and customer relations, including staple business campaigns like email marketing. Google uses it to improve their search relevance.