Artificial intelligence systems practical course tutorial. Modern problems of science and education

S.G. TOLMACHEV

ARTIFICIAL INTELLIGENCE.

NEURAL NETWORK MODELS

Ministry of Education and Science of the Russian Federation Baltic State Technical University "Voenmeh"

Department of Information Processing and Control Systems

S.G. TOLMACHEV

ARTIFICIAL INTELLIGENCE.

NEURAL NETWORK MODELS

Tutorial

St. Petersburg

UDC 004.8(075.8) Т52

Tolmachev, S.G.

T52 Artificial intelligence systems. Neural network models: textbook / S.G. Tolmachev; Balt. state tech. un-t. - St. Petersburg, 2011. 132 p.

ISBN 978-5-85546-633-1

The basic information about the structure and principles of functioning of artificial neural networks is given. The functioning of a formal neuron, the classification of neural networks according to their architecture and types of learning, typical formulations of various neural network problems and methods for their solution are considered.

It is intended for senior students studying in the specialties "Information Systems and Technologies" and "Automated Systems for Information Processing and Control".

UDC 004.8(075.8)

Peer reviewers: Dr. tech. sciences prof., head. scientific employee of OJSC "Concern "Granit-Electron"" S.N. Balls; cand. tech. sciences, prof., head. cafe I5 BSTU N.N. Smirnova

Approved by the editorial and publishing board of the university

INTRODUCTION

One of the most powerful tools for creating intelligent systems is artificial neural networks (ANNs), which model the basic information processing mechanisms inherent in the human brain. It is known that the brain works in a fundamentally different way and often more efficiently than any human-made computer. It is this fact that has been motivating scientists for many years to work on the creation and research of artificial neural networks.

The brain is an extremely complex information processing system. It has the ability to organize its structural components, called neurons, so that they can perform specific tasks (pattern recognition, sensory processing, motor functions) many times faster than today's fastest computers. An example of such a task is ordinary vision. The function of the visual system is to create a representation of the world in such a way that provides the possibility of interaction with it. The brain sequentially performs recognition tasks (for example, recognizing a familiar face in an unfamiliar environment) and spends 100–200 ms on this. Performing similar tasks of lesser complexity on a computer can take several hours.

To appreciate the enormity of the challenge of building a machine that works as perfectly as our brain, it is enough to think about some of the routine tasks that we perform every day. Suppose you are sitting at your desk, and at this time your colleague, who has returned from vacation, enters the room. He's wearing a new T-shirt, sunglasses on his tanned face, and looks a little younger because he's shaved off his beard. Do you recognize him? Undoubtedly, since disguise is not part of his plans. During the conversation, he asks you: “Where is the book that I gave you to read?”. You interpret the question as a request to return the book. Then look at your table and

you see among the books and stacks of papers lying on it the book in question, stretch out your hand to it, remove it from the pile of documents and give it to your colleague. Such everyday tasks do not require much intellectual effort from us, but the solution to each of them involves many precisely calculated steps. The complexity of solving such problems can be felt when trying to program a computer system to recognize objects by their appearance or other features, make decisions depending on the context, etc.

A simpler example is bat sonar, which is an active echolocation system. In addition to providing information about the distance to the desired object, this locator allows you to calculate object parameters such as relative speed, size of individual elements and direction of movement. To extract this information from the received signal, the tiny brain of a bat performs complex neural calculations.

What allows the brain of a person or a bat to achieve such results? At birth, the brain already has a perfect structure, allowing it to build its own rules based on what is usually called experience. Experience accumulates over time until the last days of a person's life, with especially large-scale changes occurring in the first two years of life.

The development of neurons is associated with the concept of brain plasticity - the ability to adjust the nervous system in accordance with environmental conditions. Plasticity plays the most important role in the operation of neurons as the basic information processing units in the human brain. Similarly, artificial neurons are tuned in the ANN. In general, an ANN is a machine that simulates how the brain solves a specific problem. This network is implemented using electronic components (neuroprocessors) or modeled by a program running on a digital computer. In order to achieve high performance, ANNs use a lot of interconnections between the elementary cells of calculations - neurons. Among the many definitions of neural networks, the most accurate is the definition of an ANN as an adaptive machine: artificial neural networkit's distributed

a parallel processor consisting of typical information processing elements that accumulate experimental knowledge and provide it for further processing. A neural network is similar to the brain in two ways:

1) knowledge enters the neural network from the environment

and used by the network in the learning process;

2) to accumulate knowledge, interneuronal connections are used, also called synaptic weights.

The procedure used to carry out the learning process is called the learning algorithm. Its function is to modify the synaptic weights of the ANN in a certain way so that the network acquires the necessary properties.

Weight modification is a traditional way of learning ANNs. This approach is close to the theory of adaptive linear filters that are used in control. However, for the ANN there is also the possibility of modifying its own topology, based on the fact that in the living brain, neurons can die, and new synaptic connections can be created.

Thus, ANNs realize their computing power due to their two main properties: a parallel-distributed structure and the ability to learn and generalize the knowledge gained. The generalization property is understood as the ability of an ANN to generate correct outputs for input signals that were not taken into account in the learning process (training). These two properties make ANN an information processing system capable of solving complex multidimensional problems that are currently intractable.

It should be noted that in practice, autonomous INS often cannot provide ready-made solutions. They need to be integrated into complex systems. A complex task can be divided into a number of simpler tasks, some of which can be solved by neural networks.

The areas of application of ANNs are very diverse: recognition and analysis of text and speech, semantic search, expert systems and decision support systems, stock price prediction, security systems. There are several examples of the use of ANNs in different areas.

1. Transport security systems. American firm

Science Application International Corporation has used ANN in

his project TNA. The developed device is designed to detect plastic explosives in packed luggage. The baggage is bombarded with particles that cause secondary radiation, the spectrum of which is analyzed by a neural network. The device provides a probability of detecting explosives above 97% and is capable of viewing 10 pieces of luggage per minute.

2. Neural network software packages in financial markets. The American Chemical Bank uses Neural Data's neural network system to pre-process transactions on currency exchanges, filtering out "suspicious" transactions. Citibank has been using neural network predictions since 1990. Automatic dealing shows returns that are higher than those of most brokers. It can be noted that the proceedings of the seminar "Artificial intelligence on Wall Street" make up several hefty volumes.

3. Monitoring and automatic heading of news. Location

knowledge of the topic of text messages is another example of the use of ANNs. The Convectis news server (a product of Aptex Software Inc.) provides automatic categorization of messages. By comparing the meaning of words by context, Convectis is able to recognize topics in real time and categorize the huge streams of text messages transmitted over the networks of Reuters, NBC, CBS, etc. After parsing the message, an abstract, a list of keywords and a list of headings to which this message belongs are generated.

4. Autopiloting of unmanned aerial vehicles. The LoFLYTE (Low-Observable Flight Test Experiment) hypersonic reconnaissance aircraft, a 2.5 m long jet unmanned aircraft, was developed for NASA and the US Air Force by Accurate Automation Corp. within the framework of the small innovative business support program. This is an experimental development for the study of new principles of piloting. It includes neural networks that allow the autopilot to learn by copying the pilot's piloting techniques. Over time, neural networks adopt management experience, and the speed of information processing allows you to quickly find a way out in extreme and emergency situations. The LoFLYTE is intended for supersonic flight where the pilot's reaction time may not be sufficient to adequately respond to changes in the flight regime.

Currently, ANNs are an important extension of the concept of computation. They have already made it possible to cope with a number of difficult problems and promise the creation of new programs and devices capable of solving problems that so far only a person can do. Modern neurocomputers are used mainly in the form of software products and therefore rarely use their potential of "parallelism". The era of real parallel neurocomputing will begin with the appearance on the market of hardware implementations of specialized neurochips and expansion boards designed to process speech, video, static images and other types of figurative information.

Another area of application of ANNs is their use

in specialized software robotic agents designed to process information, and not for physical work. Smart assistants should make it easier for users to interact with a computer. Their hallmark will be the desire to understand as best as possible what is required of them, by observing and analyzing the behavior of their "master". Trying to discover

in In this behavior, there are certain patterns, intelligent agents must offer their services in a timely manner to perform certain operations, such as filtering news messages, backing up documents that the user is working on, etc. That is why ANNs, capable of generalizing data and finding patterns in them, are a natural component of such software agents.

1. COMPUTERS AND THE BRAIN

1.1. biological neuron

The human nervous system can be simplified as a three-stage structure. The center of this system is the brain, which consists of a network of neurons (Fig. 1.1). It receives information, analyzes it and makes appropriate decisions. Receptors convert signals from the environment and internal organs into electrical impulses that are perceived by the neural network (brain). Receptors provide the connection of our brain with the outside world, realizing the flow of visual, auditory, gustatory, olfactory and tactile information into it. ef-

Fectors convert electrical impulses generated by the brain into output signals that control muscles, internal organs, and vessel walls. Thus, the brain controls the work of the heart, breathing, blood pressure, temperature, maintains the desired oxygen content in the blood, etc. Intermediate neurons process information received from sensory neurons and transmit it to effector neurons.

Rice. 1.1. Simplified diagram of the nervous system

It should be noted that the brain is built of two types of cells: glial cells and neurons. And although the role of glial cells seems to be quite significant, most scientists believe that the main way to understand how the brain works is by studying neurons united in a single connected network. This approach is used in the construction of artificial neural networks (ANNs).

It should be noted that there are other opinions. Some researchers believe that the main processes take place not in the neural network, but in the cells themselves, namely in their cytoskeleton, in the so-called microtubules. According to this point of view, both memory and even consciousness are determined by changes in proteins in intracellular structures and the quantum effects associated with them.

The number of neurons in the brain is estimated at 1010 ... 1011 . In a biological neuron, the following structural units can be distinguished (Fig. 1.2):

cell body (soma);

dendrites are many branching short (no more than 1 mm) nerve fibers that collect information from other neurons;

axon is the only thin long (sometimes more than a meter) nerve fiber. The axon conducts the impulse and transmits the impact to other neurons or muscle fibers. At its end, the axon also branches and forms contacts with the dendrites of other neurons;

FEDERAL AGENCY FOR EDUCATION

STATE EDUCATIONAL INSTITUTION

HIGHER PROFESSIONAL EDUCATION

"VOLGOGRAD STATE TECHNICAL UNIVERSITY"

KAMYSHINSKY TECHNOLOGICAL INSTITUTE (BRANCH)

SEI HPE "VOLGOGRAD STATE TECHNICAL UNIVERSITY"

PRACTICAL COURSE ON THE DISCIPLINE "SYSTEMS OF ARTIFICIAL INTELLIGENCE"

Educational electronic edition

Volgograd

NL - natural language

AI - artificial intelligence

LP - logic of predicates

decision maker - decision maker

MT - Turing machine

PGA - a simple genetic algorithm

PPF is a well-formed formula

PRO - primitive recursive operator

PRF - primitive recursive function

RF - recursive function

SNI - artificial intelligence system

FP - fitness function

TF - objective function

ES - expert system

INTRODUCTION

Initially, artificial intelligence was considered as the science of creating thinking machines. This area was considered the holy grail of computer science. Over time, artificial intelligence has evolved into a more pragmatic discipline. This area still includes the study of the mechanisms of thinking. Within the framework of artificial intelligence, various strategies for the computer solution of complex practical problems are considered. In addition, today it has become clear that the intellect itself is too complex an entity that cannot be described within the framework of one theory. Various theories describe it at various levels of abstraction. Learning at the lowest level is provided by neural networks that recognize machines, genetic algorithms, and other forms of computation that model the ability to adapt, perceive, and interact with the physical world. Creators of expert systems, intelligent agents, stochastic models and natural language understanding systems work at a higher level of abstraction. This level takes into account the role of social processes in the creation, transfer and extraction of knowledge. The highest level of abstraction includes logical approaches, including deductive, abductive models, truth support systems, and other forms and methods of reasoning.

This manual outlines the foundations of some low-level theories with practical tasks for the study of algorithms based on the provisions of these theories. In particular, the fundamentals of the theory of pattern recognition are considered with the task of studying linear discriminant functions and similarity functions; theory of artificial neural networks with the formulation of the problem of studying the properties of artificial neural networks on the problem of pattern recognition; genetic algorithms with the formulation of the problem of studying their properties when searching for the extremum of a function. To perform research tasks, it is necessary to be able to program in any programming language, preferably an object-oriented one.

1.1. Origins of Artificial Intelligence Theory

1.1.1. The concept of artificial intelligence

Term intelligence(intelligence) comes from the Latin intellectus, which means the mind, reason, mind, mental abilities of a person. Respectively artificial intelligence(AI, in the English equivalent: artificial intelligence, AI) is the property of automatic systems to take on individual functions of human intelligence.

Any artificial intelligence is a model of decision-making carried out by the natural intelligence of a person. Artificial intelligence can claim to be compared with natural intelligence, provided that the quality of the generated solutions is not worse than the average natural intelligence.

1.1.2. Artificial intelligence in the automation loop

In such systems, the control loop is introduced decision maker(LPR).

The decision maker has his own system of preferences regarding the criterion of object management, and even the purpose of the existence of the object. The decision maker, most often, does not agree, at least partially, with the regimes offered by the traditional ACS. The decision maker controls, as a rule, the main parameters of the system, while the rest is controlled by local control systems. There is a problem of automating the activity of decision makers in the control loop.

AI is a research direction that creates models and corresponding software tools that allow using computers to solve problems of a creative, non-computational nature, which, in the process of solving, require an appeal to semantics (the problem of meaning).

AI is a software system that simulates human thinking on a computer. To create such a system, it is necessary to study the decision maker's thinking process, highlight the main steps of this process, develop software that reproduces these steps on a computer.

1.1.3. The concept of intellectual task and activity

A feature of human intelligence is the ability to solve intellectual problems by acquiring, memorizing and purposefully transforming knowledge in the process of learning from experience and adapting to various circumstances.

Intellectual tasks- problems, the formal division of the process of finding a solution to which into separate elementary steps often turns out to be very difficult, even if their solution itself is not difficult.

The activity of the brain, aimed at solving intellectual problems, we will call thinking or intellectual activity.

Intellectual activity implies the ability to infer, generate, design a solution that is not explicitly and ready-made in the system. Derivation of solutions is possible only if there is an internal representation of knowledge in the system ( models of the outside world) - a formalized representation of knowledge about the outside world (automated subject area).

1.1.4. The first steps in the history of artificial intelligence

The first programs that implement the features of intellectual activity:

1. Machine translation (1947). In the USSR, since 1955, work in the field of machine translation has been associated with, . The task of machine translation required the separation of knowledge from code. The appearance of an intermediary language marked the first attempt to create a language for the internal representation of knowledge.

2. Automated referencing and information retrieval (1957, USA). The idea of isolating a system of connections-relationships between individual facts, embodied in the concept of thesaurus.

3. Proof of theorems (1956, USA). The emergence of a program for proving propositional logic theorems: "Logic-Theorist". In 1965, the resolution method appeared (J. Robinson, USA), in 1967, the reverse method (, USSR). Methods implement the idea of using heuristic– experimental rules for reducing the enumeration of options when deriving a solution.

4. Pattern recognition (early 60s). Ideas of recognition theory related to learning to find a decision rule on a set of positive and negative examples.

In 1956, K. Shannon, M. Minsky and J. McCarthy organized a conference in Dartmouth (USA) to summarize the practical experience of developing intellectual programs.

1.1.5. Creation of a theoretical base

In 1969, the First International Conference on Artificial Intelligence (IJCAI) was held in Washington DC. In 1976, the international journal "Artificial Intelligence" began to be published. During the 70s, the main theoretical directions of research in the field of intelligent systems were formed:

− knowledge representation, formalization of knowledge about the external environment, the creation of an internal model of the external world;

− communication, creation of languages of interaction between the system and the user;

− reasoning and planning, decision making in alternative situations;

− perception (machine vision), obtaining data from the external environment;

− training, extracting knowledge from the experience of the system functioning;

- activity, active behavior of the system based on its own goals of functioning.

1.1.6. Philosophical problems of the theory of artificial intelligence

This subsection lists the main questions and some comments on them on frequently and widely discussed problems in the theory of artificial intelligence.

Can intelligence be reproduced? Self-reproduction is theoretically possible. The fundamental possibility of automating the solution of intellectual problems with the help of a computer is provided by the property of algorithmic universality. However, one should not think that computers and robots can, in principle, solve any problems. There are algorithmically unsolvable problems.

What is the purpose of creating artificial intelligence? Let us assume that a person has managed to create an intellect that exceeds his own intellect (if not in quality, then in quantity). What will happen to humanity now? What role will the person play? Why is he needed now? And in general, is it necessary in principle to create AI? Apparently, the most acceptable answer to these questions is the concept of "intelligence amplifier".

Is it safe to create artificial intelligence? Possessing intelligence and communication capabilities many times greater than human ones, technology will become a powerful independent force capable of counteracting its creator.

1.1.7. Areas of use

1. Processing of natural languages, recognition of images, speech, signals, as well as the creation of intelligent interface models, financial forecasting, data extraction, system diagnostics, network activity monitoring, data encryption (direction - neural networks).

2. Nanotechnologies, problems of self-assembly, self-configuration and self-healing of systems consisting of many simultaneously functioning nodes, multi-agent systems and robotics (direction - evolutionary computing).

3. Hybrid control systems, image processing, tools for searching, indexing and analyzing the meaning of images, recognition and classification of images (direction - fuzzy logic).

4. Medical diagnostics, training, consulting, automatic programming, checking and analyzing the quality of programs, designing very large integrated circuits, technical diagnostics and making recommendations for equipment repair, planning in various subject areas and data analysis (direction - expert systems (ES)).

5. Transport problems, distributed computing, optimal resource loading (direction - methods of enumeration reduction).

6. Development of large software design systems, code generation, verification, testing, quality assessment, identifying the possibility of reuse, solving problems on parallel systems (direction - intellectual engineering).

7. Creation of fully automated cyberfactories.

8. Games, social behavior of human emotions, creativity.

9. Military technology.

1.2. Architecture of artificial intelligence systems

1.2.1. Elements of AIS architecture

Architecture of the artificial intelligence system(SII) - the organization of the structure within which decisions are made and knowledge is applied in a particular area. The most general scheme of the SII is shown in fig. 1. In this form, there is not a single real AIS, certain blocks may be missing. In SII, there are always only two blocks: the knowledge base and the inference engine.

Consider the main types of AIS in automated information processing and control systems:

− SII process control;

− IIS for diagnosing;

− AIS for planning and dispatching;

− intelligent robots.

https://pandia.ru/text/78/057/images/image005_133.gif" width="357" height="360 src=">

Rice. 1. Generalized SII scheme

1.2.2. Process control ISIS

The architecture of the process control ISIS is shown in fig. 2.

Features of this system:

− use of technological information for control (measured characteristics of the product about the parameters and structure of the equipment);

− the inference mechanism is used to modify the data and develop recommendations and control decisions;

− the need to work in real time;

− the need to implement temporal reasoning (taking into account changing conditions).

The work of the system is organized at three levels:

- the knowledge base (KB) includes the rules for solving problems, procedures for solving problems, data about the problem area, that is, at the level of knowledge bases, the technology itself and the entire strategy for managing the process are organized;

− working memory contains information about the specified characteristics and data about the process under consideration (DB);

− the inference mechanism (in a conventional system, this is a regulator) contains a general control mechanism for achieving the final goal (acceptable solution).

An important component is the blocks of communication between the technological process with the database and the knowledge base (blocks "Data analysis" and "Process data"). They provide the user with an upper level of access to production information about the technological process from lower-level objects, i.e. they keep the content of the database and knowledge base up to date by updating. The blocks also provide monitoring functions to prevent critical situations.

Justification and explanation of the balance and adequacy of the system's response to the development of the production situation is provided by the "Dialog interface" and "Control data" blocks.

https://pandia.ru/text/78/057/images/image007_118.gif" width="357" height="149 src=">

Rice. 2. The structure of the process control FIS

1.2.3. AIS for diagnosing

This system basically does not differ from the previous system. And since the signs of various defects may largely coincide and their manifestations may be inconsistent, then in these systems there are more detailed components of the justification and explanation of the diagnosis. Therefore, very often in such systems they introduce an assessment of decisions in terms of subjective probability.

1.2.4. AIS of robotic lines and flexible production systems

A feature of such systems is the presence of a model of the world. A robotic system operates under its own specific conditions, and in principle a detailed description of this environment is possible. This mathematical model of the environment is called model of the outside world. It is the main content of the KB of the AI robot, and the other part of the KB is the knowledge about the goals of the system (Fig. 3).

https://pandia.ru/text/78/057/images/image009_112.gif" width="294" height="100 src=">

Rice. 3. AIS of robotic lines and flexible production systems

The environment state perception system includes:

− sensors directly connected with the external environment;

− pre-processing subsystem;

− feature segmentation block;

− symbolic description of the state of the environment;

− semantic description of the state of the environment;

− block of formation of the environment state model.

The inference mechanism or behavior planning system determines the actions of the robot in the external environment as a result of the current situation and according to the global goal. Consists of:

− decision inference systems;

− block of motion planning of actuators.

The action execution system includes:

− drive control subsystem;

− drive;

− executive devices.

1.2.5. AIS planning and dispatching

Purpose: solve the problems of operational management, comparison of the results of monitoring the functioning of the object in terms of planned targets, as well as monitoring (Fig. 4).

Monitoring– continuous or periodic interpretation of signals and the issuance of messages when situations arise that require intervention.

A feature of these systems is real-time action, communication with a distributed database of an integrated control system. Such a system is necessary, since the IS data are part of the control systems.

https://pandia.ru/text/78/057/images/image011_89.gif" width="365" height="167 src=">

Rice. 4. AIS planning and dispatching

1.3. The problem of knowledge representation in AIS

1.3.1. Knowledge and Data

The problem of knowledge representation emerged as one of the problems of AI. It is associated with the creation of practically useful systems, primarily ES, used in medicine, geology, and chemistry. The creation of such systems requires intensive efforts to formalize the knowledge accumulated in the relevant science.

The term "knowledge representation" is associated with a certain stage in the development of computer software. If at the first stage programs dominated, and data played an auxiliary role of a kind of “food” for “hungry” programs, then at subsequent stages the role of data steadily increased. Their structure became more complicated: from a machine word located in one computer memory cell, there was a transition to vectors, arrays, files, lists. The crown of this development was abstract data types - classes. Consistent development of data structures has led to their qualitative change and to the transition from data representation to knowledge representation.

acquisition of knowledge

The level of knowledge representation differs from the level of data representation not only in a more complex structure, but also in essential features: interpretability, the presence of classified relationships, the presence situational relationships(simultaneity, being at the same point in space, etc., these relationships determine the situational compatibility of certain knowledge stored in memory). In addition, the level of knowledge is characterized by such features as the presence of special procedures for generalization, replenishment of knowledge available in the system, and a number of other procedures.

The presentation of data has a passive aspect: a book, a table, a memory filled with information. AI theory emphasizes the active aspect of knowledge representation: acquisition of knowledge should become an active operation that allows not only to memorize, but also to apply the perceived (acquired, assimilated) knowledge for reasoning based on it.

1.3.2. The idea of self-developing machines

Research in the field of AI arose under the influence of the ideas of cybernetics - primarily the idea of the commonality of the processes of control and transmission of information in living organisms, society and technology, in particular, in computers.

The philosophical acceptability of the AI problem in its traditional form was due to the underlying notion that the order and connection of ideas is the same as the order and connection of things. Thus, to create a structure in a computer that reproduces the “world of ideas” meant simply to create a structure isomorphic to the structure of the material world, that is, to build an “electronic model of the world”. This model was considered as a computer model - a model of human knowledge about the world. The process of human thinking was interpreted in the computer as a machine search for such transformations of the model, which were supposed to transfer the computer model to some final state. The AIS needed to know how to perform model state transformations leading to a predetermined goal - a state with certain properties. At first, there was a widespread belief in the fundamental ability of a computer to independently study the model stored in it, that is, to self-learn a strategy for achieving a set goal.

This hypothetical ability was interpreted as the possibility of machine creativity, as the basis for the creation of future "thinking machines". And although in actually developed systems the achievement of the goal was carried out on the basis of human experience with the help of algorithms based on a theoretical analysis of the created models and the results of experiments carried out on them, the ideas of building self-learning systems seemed to many the most promising. Only by the 1980s the significance of the problem of using human knowledge about reality in intellectual systems was realized, which led to the serious development of knowledge bases and methods for extracting personal knowledge of experts.

1.3.3. Reflection as a component of intellectual activity

With the development of this direction, the idea of reflexive control arose. Up to this point in cybernetics, control was considered as the transmission of signals to an object that directly affect its behavior, and the effectiveness of control was achieved using feedback - obtaining information about the reactions of the controlled object. reflexive same control- there is a transfer of information that affects the object's image of the world. Thus, the feedback turns out to be redundant - the state of the subject is known to the transmitting information, that is, the object.

Traditional AIS is based on the ideology of goal-oriented behavior such as a chess game, where the goal of both partners is to checkmate at the cost of any sacrifice. It is no coincidence that chess programs have turned out to be so important for the development of AI methods.

The analysis of the functioning of one's own model or the model of "the whole surrounding reality" (within the framework of the task set), control over its state, forecasting the state is nothing but the implementation of reflection. Reflection is a certain meta-level. With the use of high-level programming languages such as Prolog, which allows you to formulate goals and build logical conclusions about the achievability of these goals, the task of implementing reflection can already be partially solved. With their help, you can build a kind of superstructure, a kind of meta-level that allows you to evaluate the behavior of the previous one. However, when considering the term “deep reflection” or “multilevel reflection”, the problem of building models by the system itself arises. This is where abstract data types come to the rescue. They allow you to operate with data structures of any finite complexity. Thus, we can assume that artificial intelligence systems can contain a reflection model.

Thus, it is impossible to consider an intellectual system complete without the ability to evaluate, "understand" one's actions, that is, to reflect. Moreover, reflection should be considered one of the main tools for constructing the behavior of systems. In the language of mathematics, reflection is a necessary condition for the existence of an intellectual system.

1.3.4. Knowledge representation languages

In a certain sense, any computer program contains knowledge. The bubble sort program contains the programmer's knowledge of how to order the elements of a list. Understanding the essence of a computer program that solves the problem of sorting lists is not at all easy. It contains the knowledge of the programmer about the method of solving the problem, but, in addition to this knowledge, it contains others:

− how to manipulate the language constructs of the programming language being used;

− how to achieve high performance of the program;

- how to choose appropriate methods for solving particular data processing problems that nevertheless play an important role in achieving the final result, and how to organize process control.

Knowledge Representation Languages are high-level languages specifically designed to explicitly encode fragments of human knowledge, such as influence rules and a set of properties of typical objects, and the high level of the language is manifested in the fact that, as far as possible, the technical details of the knowledge representation mechanism are hidden from the user. Unlike more conventional programming languages, knowledge representation languages are exceptionally economical in terms of code size. This is largely due to the fact that the interpreter of the language takes care of a lot of little things.

Despite the noted advantages of such languages, one should not forget about the existence of certain problems in their application.

The transition from describing knowledge about the subject area in all understandable “human” language to presenting it in the form of some kind of formalism perceived by a computer requires a certain skill, since it is impossible (at least today) to describe how to mechanically perform such a transformation. Since the possibilities of inference that a program can implement are directly related to the choice of a way to represent knowledge, it is the representation of knowledge, and not their extraction, that is the bottleneck in the practice of designing ES.

The tutorial introduces readers to the history of artificial intelligence, knowledge representation models, expert systems and neural networks. The main directions and methods used in the analysis, development and implementation of intelligent systems are described. Models of knowledge representation and methods of working with them, methods of development and creation of expert systems are considered. The book will help the reader to master the skills of logical design of domain databases and programming in the ProLog language.
For students and teachers of pedagogical universities, teachers of secondary schools, gymnasiums, lyceums.

The concept of artificial intelligence.
An artificial intelligence (AI) system is a software system that simulates the process of human thinking on a computer. To create such a system, it is necessary to study the very process of thinking of a person who solves certain problems or makes decisions in a specific area, highlight the main steps of this process and develop software tools that reproduce them on a computer. Hence, AI methods involve a simple structural approach to the development of complex software decision making systems.

Artificial intelligence is a branch of informatics, the purpose of which is to develop hardware and software tools that allow a non-programmer user to set and solve their traditionally considered intellectual tasks, communicating with a computer in a limited subset of natural language.

TABLE OF CONTENTS
Chapter 1. Artificial Intelligence
1.1. Introduction to artificial intelligence systems
1.1.1. The concept of artificial intelligence
1.1.2. Artificial intelligence in Russia
1.1.3. Functional structure of the artificial intelligence system
1.2. Directions for the development of artificial intelligence
1.3. Data and knowledge. Representation of knowledge in intelligent systems
1.3.1. Data and knowledge. Basic definitions
1.3.2. Knowledge Representation Models
1.4. Expert systems
1.4.1. Structure of an expert system
1.4.2. Development and use of expert systems
1.4.3. Classification of expert systems
1.4.4. Representation of knowledge in expert systems
1.4.5. Tools for building expert systems
1.4.6. Expert system development technology
Control questions and tasks for chapter 1
Literature for Chapter 1
Chapter 2 Logic Programming
2.1. Programming Methodologies
2.1.1. Methodology of imperative programming
2.1.2. Methodology of object-oriented programming
2.1.3. Functional programming methodology
2.1.4. Logic programming methodology
2.1.5. Constraint Programming Methodology
2.1.6. Neural Network Programming Methodology
2.2. A brief introduction to predicate calculus and theorem proving
2.3. Inference Process in Prolog
2.4. Program structure in Prolog
2.4.1. Using compound objects
2.4.2. Using alternate domains
2.5. Organizing Repetitions in Prolog
2.5.1. Rollback method after failure
2.5.2. Cut and rollback method
2.5.3. simple recursion
2.5.4. Generalized Recursion Rule Method (GRR)
2.6. Lists in Prolog
2.6.1. Operations on lists
2.7. Strings in Prolog
2.7.1. Operations on strings
2.8. Files in Prolog
2.8.1. Prolog predicates for working with files
2.8.2. File domain description
2.8.3. Write to file
2.8.4. Reading from a file
2.8.5. Modifying an existing file
2.8.6. Appending to the end of an existing file
2.9. Creating Dynamic Databases in Prolog
2.9.1. Databases on Prolog
2.9.2. Dynamic Database Predicates in Prolog
2.10. Creation of expert systems
2.10.1. Structure of an expert system
2.10.2. Knowledge Representation
2.10.3. Output Methods
2.10.4. UI system
2.10.5. Rule Based Expert System
Control questions and tasks for chapter 2
Literature for chapter 2
Chapter 3 Neural Networks
3.1. Introduction to Neural Networks
3.2. Artificial neuron model
3.3. Application of neural networks
3.4. Neural network training
Control questions and tasks for chapter 3
Literature for chapter 3.

Free download e-book in a convenient format, watch and read:
Download the book Fundamentals of Artificial Intelligence, Borovskaya E.V., Davydova N.A., 2016 - fileskachat.com, fast and free download.

Tutorial « DBMS: SQL Language in Examples and Tasks by Astakhova I.F., Todstobrova A.P., Melnikova V.M., Fertikova V.V., published by FIZMATLIT in 2007 and certified by the Ministry of Education and Science, contains a selection of examples, and exercises of varying degrees of complexity to provide practical and laboratory classes on learning the basics of the SQL language within the framework of the training course dedicated to information systems with databases in the direction of training and the specialty "Applied Mathematics and Informatics". Information systems using databases are currently one of the most important areas of modern computer technology. A large part of the modern software market is associated with this area. Considering the place occupied by the SQL language in modern information technologies, its knowledge is necessary for any specialist working in this field. Therefore, its practical development is an integral part of training courses aimed at studying information systems with databases. Currently, such courses are included in the curricula of a number of university specialties. Undoubtedly, in order to provide students with the opportunity to acquire stable skills in the SQL language, the corresponding training course, in addition to theoretical acquaintance with the basics of the language, must necessarily contain a sufficiently large amount of laboratory classes on its practical use. The proposed textbook is aimed primarily at the methodological support of just this kind of classes. In this regard, it focuses on the selection of practical examples, tasks and exercises of varying degrees of complexity in compiling SQL queries, which make it possible to conduct practical classes in language learning during the academic semester.

Textbook “Artificial Intelligence Systems. Practical course” by Astakhova I.F., Chulyukov V.A., Potapov A.S., Milovskoy L.S., Kashirina I.L., Bogdanova M.V., Prosvetova Yu.V. classical university education and published by BINOM publishing houses. KNOWLEDGE LABORATORY and FIZMATLIT in 2008, prepared for lectures and laboratory classes in the disciplines "Data banks and expert systems", "Databases and expert systems", "Artificial intelligence systems", "Information intelligent systems". This book is devoted to the direction of informatics, in which in recent years there is very little domestic educational literature for higher educational institutions. Translated books are more like scientific publications than textbooks. It was necessary to come up with a lot of examples, laboratory tasks that students would perform on a computer and acquire knowledge, skills and abilities (in terms of a competency-based approach to education).

The main advantage and significant difference of this textbook from similar publications is the presence in it of about 100 examples, 235 exercises, 79 questions for repeating the material covered, 11 laboratory works in which 6 different software products are studied.

Bibliographic link

Astakhova I.F., Tolstobrov A.P., Chulyukov V.A., Potapov A.S. TUTORIALS "DBMS: SQL LANGUAGE IN EXAMPLES AND TASKS", "ARTIFICIAL INTELLIGENCE. PRACTICAL COURSE” // Modern problems of science and education. - 2009. - No. 1.;
URL: http://science-education.ru/ru/article/view?id=901 (date of access: 17.09.2019). We bring to your attention the journals published by the publishing house "Academy of Natural History"

This tutorial includes the basics of programming in the Prolog language, problem solving using the search method, probabilistic methods, the basics of neural networks, as well as the principles of knowledge representation using semantic networks. Each of the sections of the manual is provided with practical and laboratory work. The appendices contain brief descriptions of the SWI-Prolog environment, the NeuroGenetic Optimizer neural network modeling program, and the Semantic knowledge visualization program. Corresponds to the current requirements of the Federal State Educational Standard for Higher Education. For students of higher educational institutions studying in engineering and technical areas.

Book " Artificial intelligence systems. Tutorial» author Bessmertny Igor Alexandrovich was rated by the visitors of KnigoGuide and her reader rating was 0.00 out of 10.
For free viewing are provided: annotation, publication, reviews, as well as files for download.