The neural networks are computers that can calculate things that are not easy for the conventional computers: deduce way simpler of a route, understand natural language, interpret an image, deduce a sequence, etc.
The perceptron multilayered
In a very extended way of neural network and a very simple model to exemplify a neural network.
Once that choose a configuration for the neural network the only that we can do is to adjust the ω that they are adjust for the neurons.
To train a neural network need examples that know that they work and know the result.
The neural network has to adjust the ω to finish giving a very seemed result to the expected, for these has to happen the list of examples one and again until attaining the results wished, depending on the complexity these past can go of cientos to millions.
We change the image and modify the ω so that when this the cat piece the light and when it is not: it turn it off
To validate the result attained have to test with new examples in which we know the result of way to see if the network learnt of general way and no particular for the already used examples, when the network of deduce from new examples really can it to him consider trained.