Modeling the human brain?
Grasping neural networks can be a difficult thing to do, however, not impossible. Our goal with the following post is to provide you with a basic knowledge of neural networks and make you aware of the most common everyday uses.
First we will explain neural networks’ theoretical background without including complicating equations and deep mathematical frameworks. This explanation will consist of:
- how neural networks came to life
- what their purpose is
- how they work exactly.
We will finish up the post with practical examples and three use-cases that present how neural networks indirectly help your business.
The idea of neural networks
The whole idea of neural networks is inspired by the human brain’s structure. The algorithm is based on the hypothesis that learning in the human brain happens when many neurons activate at the same time.
In our previous post we discussed how with the help of predictive analytics we can recognize unexpected patterns in our historical data. The main purpose of artificial neural networks is also to learn different patterns in input data. After learning these patterns they can predict outputs for a new similar dataset.
As it can be seen in the figure below, the structure of artificial neural networks consist of 3 layers: input, hidden and output layers. These layers are built of processing neurons or so to say nodes, and they are connected by channels. The task of these nodes is to perform mathematical operations at the same time together. In general there is one input and one output layer, while there can be multiple hidden layers.
When a neuron network is learning, the neurons in the (first) input layer receive the input information, which can be a result and different features this result holds. The hidden layers perform all computations and calculations in order to transform the input data and learn these features. Each hidden layer generates an output that serves as the input for the next layer. Finally, in the output layer the network gives a prediction based on the previously learned layers of features.
The whole aim of the networks is to minimize the difference between the obtained prediction and the expected result. What is special about neural networks is that they are able to learn nonlinear features by the calculations and transformations they perform.
After the more theoretical explaination, the following example might help putting things into perspective. Let’s suppose we have an image for which we want our trained neural network to answer whether it shows a dog or a cat. By trained network we mean that previousy the network was fed images of dogs and cats, and it was told which is which animal. Based on this information our network learned how dogs and cats look like on our images.
Now, after presenting the new image, each pixel is fed to each neuron in the input layer of the network. As already explained, the neurons in every layer perform computations and feed the results to the next layers until they reach the output layer. There we will obtain whether the network classified the original image as a cat or a dog.
What can a neural network do?
Neural networks are a state of the art solution. Due to their complex but flexible algorithm they can help improve on more classical methods. They can also be well adjusted for situations where a lot of rules need to be taken into consideration.
Neural networks among others support facial recognition, real-time translations, forecasting, spam detection and also music composition. These networks are able to learn patterns in music and to generate new tunes based on that. For our participation in the AI song contest, we used neural networks for our music composition as well. Read more about that here.
Three business use-case
Now we will shortly describe three use-cases, where neural networks can contribute to impactful results within your business. Neural networks are a very strong tool for sales prediction. Based on historical sales data, neural networks can reveal unexpected demand for upcoming periods since they recognize patterns in historical records, that would otherwise not be obvious at all.
They are also very useful for operations management in terms of scheduling and planning. They can assist the scheduling of machinery, assembly lines or employee timetabling. This use-case is built on the networks’ ability to solve complex optimisation problems.
Finally, neural networks can predict business failure and success. This is a deeply researched area, where the networks take into consideration multiple factors and can detect unwanted outcomes, which can encourage buinesses to take action on time.
This post was quite on the theoretical side. However, if you would like to see an in-depth showcase about the application of a neural network in business, you can check it out here.