2. Learning

How do neural networks learn to make decisions?

Making decisions is one of the most frequently used tasks of neural networks in the field of pattern recognition and classification. In this chapter, we will first look at how yes/no decisions are made by a conventional computer program, using the Scratch programming language as an example. In comparison, we will use a neuron with one input and one output to see how this principle works with manually adjusted neural networks (a “human decision”).

In the next step, we will look at the training of this neuron using pairs of input and desired output data, which enables our network to automatically find the correct separating point (an “artificial decision”).

How networks learn to classify inputs is then shown for a neuron with two inputs and one output: After adjusting it manually so that it is able to separate two classes (a “human separation”), we will observe the automatic training of this neuron via a learning algorithm (an “artificial separation”).

Finally, we will see an example of a larger neural network with two inputs that can distinguish four classes (“multiple separations”).