What Is Backpropagation Neural Network?

Written by Coursera Staff • Updated on

In artificial intelligence, computers learn to process data through neural networks that mimic the way the human brain works. Learn more about the use of backpropagation in neural networks and why this algorithm is important.

[Featured Image] A business intelligence developer sits at her desk and analyzes data on her computer screens generated by backpropagation neural networks.

Backpropagation is an algorithm used in artificial intelligence and machine learning to train artificial neural networks through error correction. The computer learns by calculating the loss function, or the difference between the input you provided and the output it produced. When you apply backpropagation, you work backward from output nodes to input nodes to reduce the loss function and produce the desired result.

Designed to operate like natural neural networks in the human brain, neural networks are commonly used in artificial intelligence and machine learning. Jobs in these branches of computer science are some of the fastest-growing jobs in the world. The World Economic Forum expects demand for these specialists to increase by 40 percent, and the US Bureau of Labor Statistics anticipates jobs for computer scientists to grow by 23 percent through 2032 [1, 2]. If you're interested in pursuing a career in the dynamic fields of AI and machine learning, understanding how backpropagation works is a helpful step.

How does backpropagation work?

Backpropagation in a neural network helps reduce errors and improve outcomes, resulting in more reliable machine responses. It's a process of analyzing errors, comparing them to the anticipated response, and re-running the model until it produces the desired outcome. The steps (shown below) mimic how the human brain learns through trial and error.

1. Supply data to the network and give the weights time to work through the model.

2. Compare the output to the input and calculate the loss function.

3. Run the error through the network from output to input.

4. Update the weights and repeat until the system minimizes the error and makes accurate predictions with new data.

The following backpropagation example illustrates how the algorithm may work. An autocorrect feature may use deep learning to learn how you misspell the word "broccoli" when you type on a smartphone. The programmer may initially input common misspellings like "brocolli" and "broccolli."

If you misspell the word because your finger accidentally taps the key next to the one you intend to use (like "beoccoli" or "broccili"), the machine may not catch the error if the programmer did not initially provide it. Through backpropagation, it's possible to calculate the difference between the misspelling the computer expected and the one you actually made. After a series of forward and backpropagations, the computer should be able to successfully interpret "beoccoli" as "broccoli" and correct the error for you.

Types of backpropagation

The two primary types of backpropagation are static and recurrent. You use static backpropagation in a feedforward neural network, in which information moves in a single direction—from the input nodes to the output nodes. This process is useful when you want to identify characters or features for sorting and classifying. Email spam detectors and optical character recognition software typically operate on neural networks that can use static backpropagation.

Recurrent backpropagation is more dynamic because of the way recurrent neural networks operate with data sequences. Instead of moving in one direction through the network, the data becomes part of a feedback loop in the hidden nodes, allowing it to function as a memory source for future learning. This allows the network to identify patterns in the data and make predictions, which makes it useful for sentiment analysis and speech recognition. 

Who uses backpropagation?

Many companies and organizations use machine learning and deep learning systems—chatbots, intelligent assistants, and purchase recommendations—that may rely on backpropagation. You may use backpropagation in your projects if you specialize in artificial intelligence, machine learning, or data science. In these roles, you can work in various industries, such as technology, finance, health care, and retail.

Business intelligence developer

As a business intelligence developer, you create and maintain data dashboards, visualize data, and generate reports to share insights with decision-makers. For example, you may use neural networks to analyze customer satisfaction and predict how they will respond to an operational change or update to a product.

Median salary: $109,874 [3]

Computer scientist

Computer scientists create software and hardware used by computers, which requires writing and testing algorithms. In this position, you may also need to develop models to solve problems with or improve current technology. You may use backpropagation to develop architectures for neural networks and experiment with ways to optimize them.

Median salary: $176,142 [4]

Data scientist

As a data scientist, you gather and analyze data and create visualizations to present your findings. Data scientists have leveraged machine learning to uncover deeper trends, recognize patterns, and make predictions. Backpropagation can help you train the models you're using in your projects. 

Median salary: $156,864 [5]

Machine learning engineer

Machine learning engineers create programs designed to automate tasks, make predictions, and improve efficiency. As a machine learning engineer, you will likely use backpropagation to troubleshoot the algorithms behind the technology. This helps you identify and correct errors so the technology functions as intended.

Median salary: $152,434 [6]

Software developer

Software developers create programs and applications for computers, including chatbots and intelligent assistants that operate on neural networks. The creation process includes testing and troubleshooting the code to ensure it works properly. As you develop these applications, you will likely use backpropagation to identify and correct errors.

Median salary: $112,605 [7]

Pros and cons of using backpropagation

Backpropagation is an efficient and versatile method for training neural networks, which is one reason it's popular in machine learning. You can implement it even if you have a limited understanding of a neural network or its features. The model's simplicity allows you to apply it in various situations, and generally works well.

While backpropagation has proven to be a powerful tool, knowing its limitations and potential challenges in practice is important. Working through multiple iterations can make backpropagation time-consuming. At the same time, the success of this process requires high-quality data—"noisy" data can lead to irregularities.

How to get started in backpropagation

Learning how to write the code you need to test is helpful to prepare for careers that use backpropagation in machine learning and AI. Python and C++ are two popular programs in artificial intelligence and machine learning. They're both beginner-friendly programming languages used in a variety of applications. Python gives you access to several libraries used in machine learning, while C++ is a popular language for creating neural networks.

Getting started with Coursera

If you want to pursue a career in AI or machine learning, understanding the models used to train neural networks is a helpful skill. You can start by developing coding skills through a program like Python for Everybody Specialization from the University of Michigan. If you're ready to start exploring machine learning and artificial neural networks, consider the Machine Learning Specialization from Stanford. This Specialization courses cover linear and logical regressions, supervised learning, and reinforcement learning models. You can find these courses and more on Coursera.

Article sources


World Economic Forum. "Future of Jobs Report 2023, https://www3.weforum.org/docs/WEF_Future_of_Jobs_2023.pdf." Accessed January 30, 2024.

Keep reading

Updated on
Written by:

Editorial Team

Coursera’s editorial team is comprised of highly experienced professional editors, writers, and fact...

This content has been made available for informational purposes only. Learners are advised to conduct additional research to ensure that courses and other credentials pursued meet their personal, professional, and financial goals.