In order that an Artificial Intelligence can recognize and distinguish objects, it has to be “trained” first. This so-called Convolutional Neural Network (CNN) consists of several successive layers, which during the training phase learns to recognize different properties in an image and to pass this information on to the next layer in sequence. While the first layers recognise rather primitive properties such as straight lines, colours and curves, the subsequent layers specialise in more complex shapes.
In this Home-Delivery episode you will learn how the training of an Artificial Intelligence works exactly and how and why the AI is sometimes wrong in object detection.
