# What Is a Backpropagation Neural Network?

C. Daw

In the world of programming, computers and artificial intelligence, a backpropagation neural network is simply a kind of artificial neural network (ANN) that uses backpropagation. Backpropagation is a fundamental and is a commonly used algorithm that instructs an ANN how to carry out a given task. Even though this concept may seem confusing, and after looking at the equations that are required during the process seems completely foreign, this concept, along with the complete neural network, is fairly easy to understand.

For those not familiar with neural networks, an ANN, or simply an NN which stands for “neural network,” is a mathematical model that is patterned after certain features of real-life neural networks, like those found in living things. The human brain is the ultimate neural network whose functioning provides some clues on how to improve the structure and operation of artificial NNs. Like a most rudimentary brain, an ANN has a network of interconnected artificial neurons that process information.

What is fascinating about it is that an ANN can adapt and modify its structure when necessary, according to the information it receives from the environment and from within the network. It is a sophisticated computational model that uses non-linear statistical data analysis, and is capable of interpreting complex relationships among data such as inputs and outputs. It can work out problems that cannot be solved using traditional computational methods.

The idea for a backpropagation neural network first came around in the year 1969 from the work of Arthur E. Bryson and Yu-Chi Ho. In later years, other programmers and scientists refined the idea. Starting in 1974 the backpropagation neural network came to be recognized as an innovative breakthrough in the study and creation of artificial neural networks.

Neural network learning is a major task within an ANN that ensures it continues to be able to process data correctly and therefore perform its function properly. A backpropagation neural network uses a generalized form of the delta rule to enable neural network learning. This means that it makes use of a teacher that is capable of calculating the desired outputs out of the certain inputs fed into the network.

In other words, a backpropagation neural network learns by example. The programmer provides a learning model that demonstrates what the correct output would be, given a specific set of inputs. This input-output example is the teacher, or model, that other portions of the network can pattern subsequent calculations after.

The whole process proceeds methodically in measured intervals. Given a definite set of inputs, the ANN applies the computation learned from the model to come up with an initial output. It then compares this output to the originally known, expected, or good output, and makes adjustments as needed. In the process, an error value is calculated. This is then propagated back and forth through the backpropagation neural network until the best possible output is determined.

## You might also Like 