Artificial neural networks

Artificial neural networks are those that consist of special elements - neurons. They are a mathematical model of biological neurons, that is, cells that make up the human nervous system.

For the first time, neural networks were talked about in 1943, and after the invention of the Rosenblatt perceptron, the golden age came, and the networks became very popular. However, after the publication of the work of Minsk in 1969, in which the scientist proved the perceptron inefficiency under certain conditions, interest in this industry fell sharply. But the history of artificial networks does not end here. In 1985, J. Hopfield presented his research and proved that neural networks are an excellent tool for machine learning.

From biology, several concepts and principles were borrowed. The neuron is a kind of switch that receives and then transmits impulses (signals). In the event that a neuron receives a sufficiently powerful impulse, it is considered that it is activated and transmits impulses to the remaining neurons associated with it. The neuron, which remained unactivated, remains at rest, the impulse does not transmit. A neuron consists of several main components: synapses that connect the neurons to each other and receive impulses, an axon whose task is to transmit impulses, and a dendrite that receives signals from different sources. When a neuron receives a pulse above a certain threshold, it immediately transmits the signal to the next neurons.

The mathematical model is slightly different. The input of a mathematical model of a neuron is a vector that consists of a large number of components. Each of their components is one of the impulses that a neuron receives. The output of the model is one number. That is, inside the model the input vector is transformed into a scalar, which is later transferred to other neurons.

Neural networks can be taught in two ways: with the teacher and without. The learning process consists of several steps. First, an external stimulus is applied to the input of the network. Then, according to the rules, the free parameters of the neural network change, after which the network responds to input stimuli in a different way. The process must be repeated until the network decides the task. The learning algorithm with the teacher is that during the training the network already has the right answer. This method has been successfully used to solve many applied problems, but it is often criticized for being biologically implausible. Neural networks are taught without a teacher in the case when only input signals are known. On their basis, the network is gradually learning to give better output values.

The use of neural networks is really diverse. Often they are used to automate pattern recognition, forecasting, creating various expert systems, and approximating functionals. With the help of such a network it is possible to recognize sound or optical signals, to predict the exchange's indicators, to create systems capable of self-learning, which can, for example, synthesize speech according to a given text or park a car. Neural networks in the west are being used more actively, unfortunately, domestic firms have not yet adopted this technique.

Despite the advantages of ANN over conventional computations in some areas, existing neural networks are not ideal solutions. Since they are capable of learning, they may be wrong. In addition, you can not exactly guarantee that the developed neural network will be optimal. The developer is obliged to understand the nature of the problem being solved, to have a lot of information that characterizes the problem, to obtain data for testing and learning the network, to correctly choose the training method, the transfer function and the functions of the adder.