What is a neural network?

In a typical computer made according to what is called von Neumann architecture, memory banks live in an isolated module. There is only one processor that processes the instructions and rewrites memory one by one aid of serial architecture. The neural network is a different approach to calculation. In the neuron network, consisting of thousands or even millions of individual "neurons" or "knots", all processing is highly parallel and distributed. "Memories" are stored in complex connections and weights between nodes.

The neuron network is a type of computational architecture used by animal brains in nature. This is not necessarily because the neural network is an inseparable way to process than serial computer technology, but because the brain that uses serial computer technology would be much more difficult to evolve gradually. Neural networks also tend to deal with “noisy data” better than serial computers. The specialized nodes occupy information and then send the signal to the second layer on the ZAid information he received from the outside. This information is usually a binary signal "yes or no". Sometimes it moves from "no" to "yes", the knot must experience a certain threshold of excitement or stimulation.

Data moves from the input layer to secondary and tertiary layers, etc. until it reaches the final "output layer", which displays the results on the screen for analysis programmers. Human retina works on the basis of neural networks. The first level nodes detect simple geometric functions in the field of view such as colors, lines and edges. Secondary nodes begin with more abstract sophisticated functions such as movement, texture and depth. The final "output" is what our consciousness registers when we look at the field of view. Initial entry is just a complex arrangement of the fifts that would mean little without a neurological hardware that would understand this in terms of meaningful properties such asthe idea of ​​a permanent object.

In the backpropagation of neural networks, the outputs from earlier layers can return to these layers to reduce other signals. Most of our senses work like that. Initial data can cause a “educated estimate” at the final result and then look at future data in connection with this educated estimate. In optical illusions, our senses make educated estimates that have proved bad.

instead of programming neural networks algorithmically, programmers must configure a neural network with training or fine tuning of individual neurons. For example, the neural network training to recognize faces would require many training runs in which various objects "facelike" and "non -warrant" were displayed in the network, accompanied by positive or negative feedback to make the neuron network improve recognition skills.