An artificial version of our brains
Every day our brain is fed a huge amount of data from the outside world. Our brain processes this input and then converts those signals into images and sounds. The structure in our brain that takes care of this is also called a neural network. And that's what the technology Deep Learning is based on. Below, using Figure 1, we explain to you how this works.
Weight â€" how much does something weigh?
Our brains are made up of neurons "that are interconnected" and are sort of spheres that combine signals. But not all signals are equally important. Suppose a car races by and the traffic light is red, you don't cross. Then you might think that both signals (the car passing and the red light) are equally important. However, this is not true, if the light turns green and a car races by, you still should not cross. So a car racing by is clearly more important than the color of the traffic light (conversely, this is also why many people run through a red traffic light when nothing is coming). We call the degree of importance of one of the signals a weight, how heavily something weighs in the decision.
Learning by labeling
Suppose you see an image (input) consisting of thousands of tiny pixels. In the image we see a dog. As a child, you may think it is a cat or a rabbit (the prediction). You "learn" that it is not, because someone tells you so. It is given a label. In your brain, the neurons are tuned this way, to recognize a certain combination of inputs (pixels, the signals that go into the neurons) as dogs. By adjusting the weights of the neurons' inputs, you learn to recognize better and better. This happens at different stages, the layers of a neural network. First, it looks at the pixel level and aggregates, but then its outputs are also aggregated, until you arrive at 1 outcome. Often in the last layer are multiple outputs, in our example 1 for each type of animal you want to recognize.
Deep learning in practice
Now that the brain is learning to recognize, the model "is" ready for use. The advantage of an artificial digital brain is that there are no external influences that can affect the results. A computer never tires, where a human loses concentration over time and makes different, wrong decisions. This benefits the predictability and accuracy of the results!
Deep learning is behind several technologies that humans use in their daily lives. For example, innovations such as Siri, Google Translate and FaceID are based on Deep Learning. Within Datacadabra, we also make use of the technology. It allows the MowHawk to detect dirt and exotics and determine the height of grass. When various deep learning applications are combined together, even very powerful applications can be created, such as a self-driving car.
Datacadabra
We hope we have taken you into the world of Deep Learning. We at Datacadabra deploy technologies to support organizations and thereby its experts to work faster, better and more efficiently. We do this with smart, scalable solutions. Have you been inspired to also apply AI within your domain? Or do you need help finding inspiration? Then get in touch with us.