Neural network piece intelligence. Let's just say about folding: what are neural networks? Rotation of neural networks

Already richly fateful people are trying to teach the computer to think independently, but the number of pardons is too great. Chi can be repaired cei bar'єr

Try to accumulate the human intellect to squirm already as a sprat of a century. Back in 1770, the Ugrian winemaker Wolfgang von Kempelen, having created an automatic machine, which beat all those who fought with him, and brought Napoleon Bonaparte's win. For greater effect, the spectators were shown folding mechanisms of the machine. Pіznіshe vinakhіdnik buv vikritiy - the middle of the machine sidіv dosvіdcheniy shahіst, keruuchy usіm process. Prote podia caught up with a rich halas.

How everything started

A lot of years passed in the first half of the 50th century of the 20th century Frank Rosenblatt's wine perceptron is a mathematical model of connected objects, where external signals from others served as the input signal for one object.

The head of the power of the perceptron was so called the gate of the broad pardon. The whole field of elements (vuzl), po'yazanih friend with the other at the measure (matrix). A little bit of stench however reacts to the input signal - for example, they deprive it like this, like wine. But it is acceptable, we want to teach the perceptron to revise the letters one in one. We are given for entry A, B, G in the vipadical order. The model doesn't know anything about one of the objects, but then a call is given, for which the pardon function is created, which indicates that the model has been greatly pardoned. In the case of different letters, the function looks like zero (there is no pardon). In the world of that, far away, the perceptron is correct, the value of the function increases. The signal about the pardon is sent back to the input, so that the model can be corrected: the nodes at the matrix have changed their vags. As a result, the perceptron remembers its own state for the singing sets of input objects, so it starts.

The term "learning the model" is not a good idea until well known. In fact, the mill of the model came to the equal after the decal cycles of the turn-around expansion of the pardon. Likewise, the pendulum will come until jealousy after the fact that Yogo was rozgoydali. But the very fact that the model could remember its own state and get lost in the new, predicts the process of navigating a person. The model was created by the great cicava and later formed the basis of the first neural network.

A further development of the theory of neuromeasures was experienced by a numerical rage and a fall for a decade. The middle of the most famous authors was the first model of piece intelligence - Logic Theorist (LT, "logical theorist"), found in 1956 by Allen Newell and Herbert Simon, and "Eliza" - a program that supported Rozmov on English on a matter-of-fact theme, created in 1965 by Joseph Weizenbaum. However, it turned out that the computer capabilities at that time did not allow the neural networks to be blocked for more practical purposes. Zokrem, for recognizing the technique of the ymovir opponent, which was so inspired by the military.

In the 1990s, a new neurodevelopment rose up. The reach of the achievements of Data Mining (data mining), machine learning and translation of language, computer vision and other fields have been re-engineered with the latest developments in the possibilities of piece intelligence. The development of the speed of processors and the reduction of memory capacity gave them the opportunity to learn neuromeasures more quickly and more or less in real tasks. But it turned out that there were not enough given resources for training, neuromeasures were “not smart enough” - they saw a great number of pardons. Therefore, the stench knew zastosuvannya at spetsializovanih zavdannya, for example, recognizable symbols, and the burning interest to piece intelligence again fell.

Neurony people and neuromerezhі

How can neuromeres be called neurons at the head of a person, and how can you call them intellect?

The current manifestation of the process of human thought is that neurons in the cortex of the brain can communicate one with one with their axonal growths, periodically otrimuyuchi in the form of "susidiv" signals and passing from the base to the awakening camp. The brain of a human has approximately 85-86 billion neurons, and one neuron can be linked to a richness (up to 20 thousand) of other neurons. The fact that the structure is created is a thought, having attached the respect of the scientists.

The term "piece intelligence" (artificial intelligence) was first found and first voiced in 1956 by the American scientist John McCarthy. In the description of the perceptron model, more was said about how it was, but in the 50s - 60s, statistical concepts of nodes, links, and links were introduced, and the nodes were called neurons.

The result of the development of piece intelligence was an impersonal type of merezh. In our brain, the process of recognizing images, and the text, and pictures, and tse are added to the memory, are being recognized at the same time. Computers and neurotransmitters are not so universal: it happened to vigaduvat see different, leathers with some of the best attachments for your own factory: zgortkovі - it is important to vikorate for the analysis of images, recurrent ones - for the analysis of dynamic changes, auto-coders - for classification (for example, for recognizing letters, symbols), coding-decoders - for revealing the characteristics of key , important neuromeasures - for creating new objects, and special memory centers (LTSM) - for storing and collecting information.

In parallel with the types of neuromerezh, the idea of creating multi-ball neuronmes appeared, if one ball of the same neurons is the entrance to another ball. Later they began to unite different types balls in one model, and all with the method of approaching neuromeasures beyond the level of understanding to the intellect of a person.

Our days

Streamline respect to neurodevelopment and active use of the term "piece intelligence" goiters are not opposed to the results, which were far away from the transferred calculation on video cards. The skin of them to revenge richly more than the number of nuclei, lower central processor, and to instill a good memory - tse called for early learning of neuromeasures in hundreds of times against the most powerful processors. Another reason for a new turn in Galusi AI was the vindication of great obligations for education.

At the same time, the largest companies earn the world for neuromeasures: social networks (and their partners, like the collective Cambridge Analitica) target advertising for additional analytics, Nvidia and other retailers create platforms for victoria in cars-autopilots, which, as a matter of fact, recognize signs on the road. However, ІІ zastosovuyt and in the next hour zastosovuvatimut mainly for activities that are not legally required. The stench can be ahead of the appearance of an unknown individual on the object, or a shahrai-recidivist at the bank, forgive the keruvannya, but do not take responsibility for the final decision.

How can you use neuromeasures, as a basis for the development of classification of information, by piece intelligence? It is possible to understand, having learned on simple power - what can piece intelligence be used and effectively practiced without natural. Vіdpovіd - nі, to the fact that for the sake of being the leader of the person himself, there are parameters of neuromeasures for obtaining adequate results. Roughly seeming, for the perfection of the skin task, its own architecture of the neuromeasure is chosen. The situation, if it is self-sustaining, as a neuropathy to create for the accomplishment of a specific task, is still not visible on the horizon.

Irrespective of those who could still beat a person in checks, at once, like a squirt of money for a leather machine that beat a person, people are happy, what they did.

Neuroinformatics and neurocybernetics are one of the directions of piece intelligence. Oskilki " piece intelligence - this is straight forward informatics, a method of which is the development of software and hardware features that allow you to overcome traditional intellectual tasks", and before these tasks include the creation of intelligent robots, optimal control, learning and self-learning, recognition of images, forecasting, etc. - the same task, for the accomplishment of such neuron networks and the most widely.

As in the case of the creation of classical expert systems, a person-expert (possibly, with the help of an engineer for knowledge) is obliged to formalize his knowledge (think of it as a natural set of rules or templates), Neuromedial expert systems self-learn from experimental data (facts) fixes. We can hope for the creation of neuroexpert systems for the capacity of a human expert, for example, for a new area of activity that is emerging, de needing diagnostics.

Neuromaneous algorithms are successfully developed for the implementation of folding practical tasks, as they are traditionally considered to be intelligent: recognition of objects (that other task of recognition of images and objects in the image), control of unmanned aircraft, medical yards.

Obviously, technologies and methods of piece intelligence to stun the main voice on the situation, which allow one or the other dekilkom such features:

Algorithm of variation of nevidomy chi can be vikoristany through the exchange of resources of the computer,
the task cannot be assigned in numerical form,
The objectives of the task cannot be expressed in terms of a precisely assigned objective function-criterion.

However, shards knowledge - information is formalized, so that the process of logical vysnovka is won", then we can say that neuromechanism is taking facti(actual knowledge about the world, presented at a glance at the initial selection) and the process of learning the form regulations− knowledge of how to describe the knowledge by the neuromeasurement method of realization. Rules for adopting a decision can be taken from a neural network and written in one of the formalisms of representation of knowledge traditional for classical expert systems (for example, in a set of production rules of a logical pattern). But you can just rely on prompting the neuromechanical manifestations of the algorithm to accept the solution, so changing the interpretation of yoga is less important than the possibility of solving the problem by the way.

Possibility of the Swedish learning and further learning of the neuromechanical expert systems allows them to display the peculiarities of the subtle hello world and operate with current knowledge, as well as the traditional way of formalizing the knowledge of people-experts, trival and laborious.

The stench was divided into blocks of natural neural networks of the human nervous system.

Piece neural networks

The culprit of the first neurocomputer, Dr. Robert Hecht-Nielsen, gave such an understanding of the neural network: "A neural network is a system that is composed of a series of simple, highly interrelated processing elements, like processing information in a way of dynamic response to calls" .

Basic structure of piece neural networks (INS)

The idea of INS is based on reconciliation, which can imitate the human brain to the robot, having created the necessary connections for auxiliary silicon and wires, such as in living neurons and dendrites.

The human brain is made up of 100 billion nerve cells, called neurons. The stench is tied to less than a thousand people Akson. Podrazniki іz ovnіshny sredovishcha or signals vіd organіv sensitively accepted dendrites. These input signals make electrical impulses, as if they move easily through the neuropath. Then a neuron can help to revitalize other neurons, yakі can nadіslati tse podomlennya dalі or zagalі її not correct.

Pieces of neuron networks are composed of a number of nodes that mimic the biological neurons of the human brain. Neurons connect with each other and interact one by one. Vuzli can accept input data and perform the simplest operations on data. After these operations, data are transferred to other neurons. The findings of a skin node are called yoga activation.

The leather lanka is tied with a wag. ІNS buildings before the start, as if by way of changing the value of the vag. A simple INS is shown on the stepping baby:

Tipi piece neural networks

Є two types of piece neural netting topologies - with a direct connection and with a ringing sound.

The flow of information is single-directed. The block transmits information to other units, which do not take any of the same information. There are no loops of a salutary link. The stench may be fixed, come in and out.

Here loops of a zvorotny zvyazku are allowed.

How to process piece neuron arrays

The topology shows the scheme, the skin arrow is the link between two neurons and indicates the path for the flow of information. The skin ligament may be a number, a number, as a control signal between two neurons.

As I create a “good” and “needed” life, there is no need to trim the vag. However, as if creating a network of "disgraceful" or "non-bad" whiskers, or a pardon, then the system corrects its own coefficients for a plethora of offensive results.

Machine learning for piece neural networks

ІNS building before training, and the stench may be training. Іsnuє kіlka strategіy navchannya

Navchannya - includes a teacher in himself, who gives in a measure the initial choice on how the reader knows the advice. Measures to improve your results with the teacher's opinions and correct your own coefficients.

A teacher without a teacher is necessary, if there is no primary selection with the necessary guidance. For example, at the heads of clustering, tobto. subdivided many elements into groups according to the same criteria.

Learning from support is the whole strategy, prompted by caution. Merezha takes decisions, guarding for her honours. As a more cautious negative, shaping one's own vagi, so that the mother can work for various necessary decisions.

Algorithm

Bayesivsky Merezhi (BS)

Qi graphic structures for the representation of imovirnіsnyh vіdnosin between a set of vipadkovyh change.

In these chains, the leather vuzol is a vipadkovy change with specific propositions. For example, in medical diagnostics, the vuzol Cancer is a proposition that the patient has cancer.

The ribs that connect the knots are movirnіsnі fallows between tsimi vypadkovymi values. If two vuzliv one pours into another vuzol, then the stench is guilty of buti po'yazani without any middle. The strength of the tie between the changes is clearly determined by the imovirnistyu, which is connected with the skin knot.

If you are only fouling on the arcs in BN, you cannot turn back to the knot just following the straight line of the arc. Zvidsi BNZ is called an acyclic graph.

The structure of the BN is ideal for acquiring knowledge and data that are guarded. BN can be found out, in order to recognize the causal-hereditary links and to understand the different problems and to transfer the future, to inspire at the time of the given data.

Dewiring neuron networks

The stench of the building vykonuvati zavdannya, yakі simple for people, ale twist for cars:

Aerospace - aircraft autopilot;

Cars automotive systems guidance;

Військові - support meti, autopilot, signal/image recognition;

Electronics - forecasting, failure analysis, machine zir, voice synthesis;

Finance - assessment of non-viability, credit consultants, mortgage, portfolio of a trading company and others.

Signal processing — neuron arrays can be trained to process an audio signal.

Piece intelligence, neural networks, machine learning - what do these popular understandings really mean? For most uninitiated people, like myself, the stench always seemed fantastic, but in reality the essence of them lies on the surface. For a long time, I had an idea to write my simple one about piece neuron meshes. Find out for yourself and tell others about what technology is, how it works, look at its history and prospects. In this article, I tried not to climb into the net, but simply to popularize the prospects directly in the world of high technologies.

Piece intelligence, neural networks, machine learning - what do these popular understandings really mean? For most uninitiated people, like me, the stench always seemed fantastic, but in reality the essence of them lies on the surface. For a long time, I had an idea to write my simple one about piece neuron meshes. Find out for yourself and tell others about what this technology is, how it works, look at its history and prospects. In this article, I tried not to climb into the net, but simply to popularize the prospects directly in the world of high technologies.

Trochs of history

For the first time, the understanding of the piece-wise neural networks was viniklo when trying to model the process of the brain. The first serious breakthrough in this sphere is the creation of the McCulloch-Pitts model of neural networks in 1943. For the first time, the model of a piece neuron was broken up. They also proponated the construction of a number of elements for the selection of logical operations. Ale, the head of the head, it was brought to light that such a fence was built up.

The coming important crop was the development by Donald Hebb of the first calculation algorithm for the INS in 1949, which became the main one for the coming decades. In 1958, the father of Frank Rosenblatt, the parteptron, was a system that mimics the processes of the brain. In its time, the technology is not small analogues and dosi є the main one in neural networks. In 1986, practically one hour, independently, one type of one American and Radianian research was completely supplemented by the main method of training a bagatospheric perceptron. In 2007, roci neuron networks were transferred to another people. The British computer scientist Geoffrey Hinton first developed an algorithm for the deep learning of rich-ball neural networks, which, for example, is victorious for robotic unmanned vehicles.

Briefly about the smut

In the wild sense of the word, neuron networks are mathematical models that are based on the principle of the network of nerve cells of a creature's organism. INS can be implemented both in programming and in hardware solutions. For simplicity, the neuron can be spawned, as if it were the middle, as if it were an impersonal opening, and one day. In some order, numerical input signals are formed at output signals, which determines the calculation algorithm itself. Differences are given to the skin input of a neuron, which then expands to interneuronal connections (synopses). The synapse has one parameter - vaga, so that the input information changes when passing from one neuron to the next. Easier principle of work and neuromerezh can be represented on the example of mixing colors. Blue, green, and red neuron are different vagaries. Information of the neuron, which will more dominate in the offensive neuron.

Neuromerezha itself is a system of impersonal such neurons (processors). Okremo qi processors are easy to do (more simple, lower processor of a personal computer), but being enrolled in great system the neurons of the building are more foldable.

The fallow in the area of stasis of the neuromeasure can be interpreted in different ways, for example, from the point of view of the machine learning of the INS - the method of recognizing the images. From the mathematical point of view, it is a richly parametric task. From the point of view of cybernetics - a model of adaptive control of robotics. For piece intelligence, INS is the main warehouse for modeling natural intelligence for additional computational algorithms.

The main advantage of neuromeasurement over the most significant calculation algorithms is the possibility of learning. In general sense, the words are learned from the most important coefficients of communication between neurons, as well as in the contextual data and the revealed folded deposits between input and output signals. In fact, in the distance, the beginning of the neurotransmitter means that the system will show the correct result on the data input, like in the beginning of the day in the initial selection.

Today's camp

I would not have been such a rich technology, while INS is still far away from the possibilities of the human brain that myslennia. Prote, already at the same time neurotransmitters zastosovuyut in rich areas of activity. As long as the stench is not accepted in the future, high-intellectual solutions, but they can replace the people there, where earlier it was necessary. Among the numerical areas of INS can be identified: the creation of systems of vibrational processes that are self-guided, unmanned transport systems, image recognition systems, intelligent security systems, robotics, systems for monitoring viability, voice interfaces in interfacing modes. Such a wide extension of the neuron is also zoomed in by the appearance different ways early education INS.

On this day, the market of neural networks of greatness is worth millions and millions of dollars. As practice shows, most of the technologies of neuromeasures in the whole world are little explored one in one. Proteza zastosuvannya neuromerezh - tse even more costly employment, as in the case of more vipadkіv can afford to own only great companies. For the development, training and testing of neural networks, it is necessary to have a large number of pressures, obviously, there is plenty of it in abundance for the IT market. Among the main companies that conduct research in this gallery, one can name Google DeepMind, Microsoft Research, IBM, Facebook and Baidu.

Obviously, everything is good: neurotransmitters are developing, the market is growing, but for the time being, the big task has not been done. Humanity did not dare to create technology, if only I could get closer to the human brain. Let's take a look at the main insights between the human brain and the pieced neuromeasures.

Why are neurotransmitters so far from the human brain?

The most important authority, which fundamentally changes the principle and efficiency of the robotic system - value transfer signals in piece neural networks and in biological networks of neurons. On the right, in the fact that in the INS neurons transmit values, which are real values, that is, numbers. The human brain is capable of transmitting impulses with a fixed amplitude, and these impulses are practically mitigated. Zvіdsi viplivaє low transcendence of the human network of neurons.

First, the lines of communication in the brain are richly effective and economical, lower in the INS. In another way, the impulse circuit ensures the simplicity of the implementation of technology: there is enough choice of analog circuits instead of folding calculation mechanisms. Zreshtoy, impulse measures of protection from sound shifts. The diysny numbers are up to a splash of noise, in the wake of which the imovirnіnіnіnіnіnі pardon is raised.

Podbag

Insanely, the remaining decades have become a boom in the development of neural networks. We have been told that it is because the process of training the INS has become richly swedish and simple. So they began to actively develop so-called “achieved” neurotransmitters, as they allow to speed up the process of introducing technology. It is still too early to talk about those who can, if they are neuromeasures, fully confirm the ability of the human brain, and the ability of the next ten years of the INS to replace a person on the fourth of the most important professions of the Daedal becomes more similar to the truth.

For those who want to know more

The Great Neural War: What is Google Really Starting
How cognitive computers can change our future

Scientific and practical conference

"Croc at the Future"

Andriychuk Andriy student 9 class A

MBOU "ZOSH No. 47" m. Chiti

Scientific curator: Mikhailov Y.I. teacher of informatics and physics, MBOU "ZOSh No. 47" m.Chiti (higher category)

metro Chita - 2018

Neuron networks. Piece intelligence

Andriychuk Andriy

MBOU "ZOSh No. 47"

9 class "A"

Short instruction

Under the hour of operation of the hard computer systems, Kozhen asked: “And how can a machine think and behave like a person? ".
In this rank, the development of II began to create a similar intelligence in machines, similar to a human one.

Neuroinformatics and neurocybernetics are one of the directions of piece intelligence. Shards of "piece intelligence - directly informatics, a method of which is the development of software and hardware devices that allow you to overcome traditionally intellectual tasks", and before these tasks include the creation of intelligent robots, optimal control, self-management. - that is the purpose of the task, for the development of such neuron networks and the most widely.

Neuron networks. Piece intelligence

Andriychuk Andriy

Russia, Trans-Baikal Territory, Chita

MBOU "ZOSh No. 47"

9 class "A"

Abstract

Neural network algorithms are successfully, traditionally considered intelligent: (the other tasks and objects in the image), control of unmanned aerial vehicles, just.

Obviously, technologies and methods of piece intelligence to stun the main voice on the situation, which allow one or the other dekilkom such features:

the algorithm of the solution is unknown and can be hacked through the exchange of resources of the computer;

task cannot be assigned to the numerical form;

The objectives of the task cannot be expressed in terms of a precisely assigned objective function-criterion.

However, the fragments of "knowledge - the information is formalized, like vikoristovyut in the process of logical vysnovka", then we can say that the neuromeasurement takes facts (actual knowledge about the world, presented at the sight of the primary vibrator) and in the process neuromeasurement method of rozvyazannya. Rules for adopting a decision can be taken from a neural network and written in one of the formalisms of representation of knowledge traditional for classical expert systems (for example, in a set of production rules of a logical pattern). But you can just rely on prompting the neuromechanical manifestations of the algorithm to accept the solution, so changing the interpretation of yoga is less important than the possibility of solving the problem by the way.

Possibility of modern learning and further development of neuromechanical expert systems allows them to reflect the peculiarities of the outer world, which is changing quickly, and to operate with current knowledge, just like the traditional way of formalizing the knowledge of people-experts and more difficult.

Actuality doslіdzhen connected with neural networks are connected with the fact that the processing of information, which is in the human brain, is examined by the methods of digital processing. The human brain works like a foldable, non-linear, parallel numbering attachment.

It has been scientifically proven that the brain is made up of a majestic number of nerve clitins (neurons). The number of structural connections in the human brain, created by more than one neuron, varies from a dozen to a hundred thousand. In this way, a neuron network is created, in which nerve signals pass. Qi signals є the cause of the change will become neuroniv and їх spoluk. Depending on the number of nerve signals, the activity of the brain should be deposited.

Tsіl given robotics - tse creation that far away completely to the prototype of neural networks(Add your programs with urakhuvannyam)

Meta includedzavdan :

Analysis of the actual information from this directly;

Development and adjustment programs;

Zastosuvannya and demonstration by law and principles of neural networks and piece intelligence;

Conducted research and experiments.

Follow-up object: piece intelligence;

Subject of inquiry: neural network - one of the ways to implement piece intelligence;

follow-up methods:

- follow-up directly related to piece intelligence, machine learning and neuromeasures;

zagalnennya otrimanih danikh;

Experimental follow-up models of neural networks;

Approbation software for the model of neural networksadd the name, indicate the language of the program(Navchannya, rozpіznannya).

Hypothesis: at the basis of the working hypothesis lies the assumption about those that are used, neural networks and piece intelligence can significantly improve the life of a person, as if alive in the information society, become at the service of modern information technologies.

Scientific novelty: A new model of piecewise neural networks was developed, which allows describing signal processing algorithms in terms of elements and links between them. The creation of the model add the name of the program

Practical significance

The creations of the robotic process and the software package can be developed for description, compilation, visualization, optimization and launch of neural networks within the framework of the new model. With any expansion, the algorithms and architecture allow the implementation of other models of neural networks.

The work is made up of three divisions. The first one has a clear understanding of neural networks, piece intelligence. The history of neural networks and piece intelligence is presented, classification is made.

Another branch has the technology of distribution software complex, reviewed the main methods of work.

In the third section, a description of the results of the modeling is presented, an analysis of the experiments is carried out.

Neuron networks. Piece intelligence

Andriychuk Andriy

Russia, Trans-Baikal Territory, Chita

MBOU "ZOSh No. 47"

9 class "A"

follow-up plan

Designation of the problem and food for follow-up;

Methods of follow-up: selection of the core information and analysis of taking data from the selected nutrition of the follow-up, revealing the main features of the follow-up nutrition, designing the key ones to understand the follow-up.

Hid date:

Statement by those doslіdzhennya; revealing the relevance of the research;
Designated stake food, tied with the theme of the study;
Iyavlennya dzherel іnformatsiї, nebhіdnі conducted dolіdzhennya;
Vyvchennya dzherel іnformatsiї; revealing basic understanding, terms;
Analysis and analysis of the collected information, selection of the main categories of research;
Rozrobka that folded to the plan of the last work;
Processing and analysis of information;
Written doslidnitskoy work;
The plan is drawn up in order.
Zakhist at the school conference.

Neuron networks. Piece intelligence

Andriychuk Andriy

Russia, Trans-Baikal Territory, Chita

MBOU "ZOSh No. 47"

9 class "A"

Zmist

Entry;

main part;

The last part;

Visnovok;

List of gerel information.

Neuron networks. Piece intelligence

Andriychuk Andriy

Russia, Trans-Baikal Territory, Chita

MBOU "ZOSh No. 47"

9 class "A"

Entry

Intellect - it’s even more wildly rozum’s building, as it includes the ability to work visnovka, plan, solve problems, think abstractly, think folded ideas, learn quickly and learn on the basis of information.

The intellect of a person is an organic structure, for all pluses and minuses.

Neuronna merezha - one of the ways to implement piece intelligence (II).
The rozrobtsi ІІ іsnuє has a large area - machine learning. You will learn the methods of inspiring algorithms, building independently. It is necessary, as it does not have a clear vision of whether it is a task. For this reason, it is easier not to guess the right solution, but to create a mechanism, which itself will foresee a method for joking.

neurositka model the work of the human nervous system, in particular, building up to self-management with improved forward access. In this rank, at once, the system will give less and less pardons.

main part

History of neural networks

The term "neuronal mesh" appeared in the middle of the 20th century. The first robots, from which the main results were taken away from them directly, were broken up by McCulloch and Pitts. In 1943, they developed a computer model of the neural network on the basis of mathematical algorithms and the theory of the activity of the brain. The stench hung down, which neurons can be simply looked at as an attachment, which can be operated on with two numbers, and they called this model threshold logic. Similar to their biological prototype of the McCulloch-Pitts neuron, they began to learn how to build parameters that describe synaptic conduction. The successors proposed the construction of a network of electronic neurons and showed that such a network can be practically implemented, whether numerically or logically, operations. McCulloch and Pitts let it go, that such a line can also be learned, recognizable, understand, to be able to draw all intelligence.

This model laid the foundations for two different approaches to the follow-up of neural networks. One approach is to guide the development of biological processes in the brain;

in 1949, the Canadian physiologist and psychologist Hebb, who came up with ideas about the nature of brain neurons and their interaction;

In 1954, at the Massachusetts Institute of Technology, using Farley and Clark's computers, they developed an imitation of Hebb's string. Also, studies of neural networks for additional computer modeling were carried out by Rochester, Holland, Hebit and Duda in 1956;

In 1957, Rosenblatt developed a mathematical computer model for the perception of information by the brain based on a two-ball neural network, which is being developed. Starting hour merezha is given victorious arithmetic dії adding that vіdnіmannya;

Interest in the study of neural networks after the publication of works from the machine learning of Minsky and Papert in 1969. They have revealed the main enumeration problems that are to be blamed for the computer implementation of piecewise neural networks;

one of the most important steps, which stimulated further research, was the development in 1975 by Verbos of the method of reversible extension of the pardon, which allowed the effective implementation of the task of training bagatospheric meshes and the problem of “folding modulo 2”;

1975 Fukushimoy's fate was the splitting of the cognitron, which became one of the first rich-ball neural networks. The actual structure of the merezhі and the methods that are victorious in the cognitron for setting up the external voices of the links varied from one strategy to another. Skin zі strategіy is small svoіvagi that nedolіki;

the algorithm of parallel split processing of data in the mid-1980s became popular under the name of connectivism. In 1986, Rummelhart and McClelland's robots developed the connection of a vicorous steel for computer simulation of neuronal processes.

Piece neural networks

The culprit of the first neurocomputer, Dr. Robert Hecht-Nielsen, gave such an understanding of the neural network: “A neural network is a system that is composed of a series of simple, highly interrelated processing elements, like processing information in a way of dynamic response to calls.”

Basic structure of piece neural networks (INS)

Tipi piece neural networks

Є two types of piece neural netting topologies - with a direct link and a back link.

Yak vlashtovana neuron mesh

The neural network is a model of the robot of the human nervous system, especially the ability to self-learn with the improvement of the forward access. In this rank, at once, the system will give less and less pardons.

Just like our nervous system, the neuromeasurement is made up of a few counting elements - neurons, like stitching on a few balls. Dani who need to go to the entrance of the neuromeasure, go through the subsequent processing of the skin ball of the merezh. In each skin, the neuron has several parameters, which can change indefinitely depending on the subtraction of the results - in whom it is necessary to change the beginning of the measure.

Let's say that the task of neuromeasures is to revitalize cats from dogs. A great array of signed images of cats and dogs is submitted for the creation of a neural network. The neural network analyzes the signs (zonal lines, shapes, their dimensions and colors) from these pictures and the future such a recognition model, as well as minimizing the number of pardons and standard results.

A little lower than the representations, the process of robotic neuromeasurement, the task of recognizing the number of the postal code, is written in the form of a hand.

Figure 1 Attachment of the neural network

Neuronal networks with direct connection

Malyunok 2 Neuronal links from direct link

Figure 3 Neuronal networks from the gate link

Machine learning for piece neural networks

ІNS building before training, and the stench may be training. Іsnuє kіlka strategіy navchannya

Navchannya - includes a teacher in himself, who gives in the measure of the initial selection, on which the reader knows the advice. Measures to improve your results with the teacher's opinions and correct your own coefficients.

A teacher without a teacher is necessary, if there is no primary selection with leading recommendations. For example, at the heads of clustering, tobto. subdivided many elements into groups according to the same criteria.

Learning from reinforcements - this strategy, prompted by caution. Merezha takes decisions, guarding for her honours. As a more cautious negative, shaping one's own vagi, so that the mother can work for various necessary decisions.

Bayesivsky Merezhi (BS)

Qi graphic structures for the representation of imovirnіsnyh vіdnosin between a set of vipadkovyh change.

In these chains, the leather vuzol is a vipadkovy change with specific propositions. For example, in medical diagnostics, the vuzol Cancer is a proposition that the patient has cancer.

If you are only fouling on the arcs in BN, you cannot turn back to the knot just following the straight line of the arc. Zvidsi BNZ is called an acyclic graph.

Popularity of neural networks

Until 2010, the rock simply did not have a data base that was large enough to learn the neurotransmitters to play the songs of the day, mainly due to the recognition of that classification of the image. That neuromerezhi often had mercy: they strayed the intestines with the dog, otherwise, even worse, a sign of a healthy organ with a prominent organ, affected by plumpness.

In addition, in 2010, the ImageNet database appeared, which will cover 15 million images in 22,000 categories. ImageNet has extensively reviewed the data bases of the images and was available to any previous source. With such commitments to these neuromeasures, it is possible to take a practically unforgiving decision.

Prior to this path, the development of neuromeasures had another, but less important, problem: the traditional method of training was ineffective. Irrespective of those who play an important role in the number of balls in the neural network, the method of creating the network is also important. Previously, the reverse ciphering method could effectively learn only the remaining version of the merezhі. The process of starting a letter is too trivial for practical zastosuvannya, and attaching the balls of deep neuromerezh did not function properly.

In 2006, the results of the most important problem in 2006 were reached by three independent groups of scientists. Firstly, Geoffrey Hinton, having realized the beginning of the measure for help, just like a leather ball of okremo. In a different way, Yan LeKan uttered a song to solve the problems of image recognition. Nareshti, Joshua Bengio having developed a cascade autoencoder, which allowed the insertion of the mustache balls in the deep neural network.

Successful planting of neural networks

Table 1

The knowledge of the piece intellect marked the risk of cardiological diseases as effective for real doctors.

Finance

The Japanese insurance company Fukoku Mutual Life Insurance has placed a contract with IBM. According to youmu,34 tips from a Japanese company to replace the IBM Watson Explorer AI system.

Business

piece intelligencethe essence of painting recommendation mechanisms in online stores and services.Algorithm Yandex Data Factorybuilding transfer inflow of shares.Neural networks that analyze natural language,You can tweet for the creation of chatbots.

Transport

unmanned vehicles - a concept that most of the great concerns are working on, point out thatsmart cars to be on the roads until 2025.

Promislovist

technology of synthetic molecules, steel melting, warehouse processing.

Silsk state

determination of the optimal time for inspection and harvesting of agricultural crops.

art

photo and video edits, neuron companies have already recorded two albums, the Japanese algorithm has written a book"The day when Computer wrote a novel",the program beat the strongest grave in the world

Bezpeka

look for sharks in coastal waters and people on the beaches

Practical part

Insert screenshots report description robotic program process!

Visnovok

Neuronal networks, the technology of the middle of the last century at once changes the work of whole galleys. The reaction of the suspense is ambiguous: some of the possibilities of neuromeasures call out to the capture, and others - to hesitate to hesitate in their bark like fahivtsiv.

Prote do not scratch, where the machine trains come, there are people. Just because neuromeasures make diagnoses better for a living doctor, it doesn’t mean that in the future we will be treated all the way. Nayimovіrnіshe, the drug is pracyuvatime at once from the neuromeasure. Similarly, the IBM Deep Blue supercomputer won the checks from Garry Kasparov in 1997, but the people from the checks did not go anywhere, and the famous grand masters of the dosit were spent on the lining of glossy magazines.

The piece intelligence of the future of all mankind, the development of neural networks, is the majestic crop of the future, where the physical work of the people will not be needed.

Already at the same time mi bachimo, naskіlki shvidko and bezpomilkovo computer vykonuє dії, as if not under the force of dozens of scientists. Investing and developing technologies II, people develop overnight in a thousand spheres, pieces of piece intelligence can be practically stagnant in all spheres of life.

List of victorious literature

Categories
- Android
- Programs
- iOS
- Printers
- Editors
- Games
- SmartTV
- Internet
- Gadgets
- Zalizo

more popular