Standardisation of indicators of quality of products cognitive technologies
The standard set of quality indicators, reflecting the principal features of functioning and development of intellectual systems is proposed. Comparative assessment of some of them is given.
Gespeichert in:
| Datum: | 2014 |
|---|---|
| 1. Verfasser: | |
| Format: | Artikel |
| Sprache: | English |
| Veröffentlicht: |
Інститут проблем штучного інтелекту МОН України та НАН України
2014
|
| Schriftenreihe: | Искусственный интеллект |
| Schlagworte: | |
| Online Zugang: | http://dspace.nbuv.gov.ua/handle/123456789/85302 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Zitieren: | Standardisation of indicators of quality of products cognitive technologies / S.S. Antsyferov // Искусственный интеллект. — 2014. — № 3. — С. 72–79. — Бібліогр.: 14 назв. — англ. |
Institution
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-85302 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-853022015-07-25T03:01:46Z Standardisation of indicators of quality of products cognitive technologies Antsyferov, S.S. Интеллектуальные системы планирования, управления, моделирования и принятия решений The standard set of quality indicators, reflecting the principal features of functioning and development of intellectual systems is proposed. Comparative assessment of some of them is given. Запропонована стандартна сукупність показників якості, що відбиває принципові особливості функціонування і розвитку інтелектуальних систем. Дана порівняльна оцінка деяких з них. Предложена стандартная совокупность показателей качества, отражающая принципиальные особенности функционирования и развития интеллектуальных систем. Дана сравнительная оценка некоторых из них. 2014 Article Standardisation of indicators of quality of products cognitive technologies / S.S. Antsyferov // Искусственный интеллект. — 2014. — № 3. — С. 72–79. — Бібліогр.: 14 назв. — англ. 1561-5359 http://dspace.nbuv.gov.ua/handle/123456789/85302 681.518.3 en Искусственный интеллект Інститут проблем штучного інтелекту МОН України та НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| topic |
Интеллектуальные системы планирования, управления, моделирования и принятия решений Интеллектуальные системы планирования, управления, моделирования и принятия решений |
| spellingShingle |
Интеллектуальные системы планирования, управления, моделирования и принятия решений Интеллектуальные системы планирования, управления, моделирования и принятия решений Antsyferov, S.S. Standardisation of indicators of quality of products cognitive technologies Искусственный интеллект |
| description |
The standard set of quality indicators, reflecting the principal features of functioning and development of
intellectual systems is proposed. Comparative assessment of some of them is given. |
| format |
Article |
| author |
Antsyferov, S.S. |
| author_facet |
Antsyferov, S.S. |
| author_sort |
Antsyferov, S.S. |
| title |
Standardisation of indicators of quality of products cognitive technologies |
| title_short |
Standardisation of indicators of quality of products cognitive technologies |
| title_full |
Standardisation of indicators of quality of products cognitive technologies |
| title_fullStr |
Standardisation of indicators of quality of products cognitive technologies |
| title_full_unstemmed |
Standardisation of indicators of quality of products cognitive technologies |
| title_sort |
standardisation of indicators of quality of products cognitive technologies |
| publisher |
Інститут проблем штучного інтелекту МОН України та НАН України |
| publishDate |
2014 |
| topic_facet |
Интеллектуальные системы планирования, управления, моделирования и принятия решений |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/85302 |
| citation_txt |
Standardisation of indicators of quality of products cognitive technologies / S.S. Antsyferov // Искусственный интеллект. — 2014. — № 3. — С. 72–79. — Бібліогр.: 14 назв. — англ. |
| series |
Искусственный интеллект |
| work_keys_str_mv |
AT antsyferovss standardisationofindicatorsofqualityofproductscognitivetechnologies |
| first_indexed |
2025-07-06T12:30:47Z |
| last_indexed |
2025-07-06T12:30:47Z |
| _version_ |
1836900729872187392 |
| fulltext |
ISSN 1561-5359 «Искусственный интеллект» 2014 № 2 72
4А
УДК 681.518.3
S.S. Antsyferov
Moscow State Technical University Mirea, Russia
Russia, 119454, c. Moscow, Vernadsky ave., 78
Standardisation of Indicators
of Quality of Products Cognitive Technologies
С.С. Анцыферов
Московский государственный технический университет МИРЭА, Россия
Россия, 119454, г. Москва, пр. Вернадского, 78
Стандартизация показателей качества продукции
когнитивных технологий
С.С. Анциферов
Московський державний технічний університет МІРЕА, Росія
Росія, 119454, м. Москва, пр. Вернадського, 78
Стандартизація показників якості продукції
когнітивних технологій
The standard set of quality indicators, reflecting the principal features of functioning and development of
intellectual systems is proposed. Comparative assessment of some of them is given.
Key words: indicators of quality, cognitive technologies, intellectual system, neural network,
probability, interpolation, stability, structure, processing speed, reliability.
Предложена стандартная совокупность показателей качества, отражающая принципиальные особенности
функционирования и развития интеллектуальных систем. Дана сравнительная оценка некоторых из них.
Ключевые слова: показатели качества, когнитивные технологии,
интеллектуальная система, нейронная сеть, вероятность, интерполяция, устойчивость,
структура, быстродействие, надежность.
Запропонована стандартна сукупність показників якості, що відбиває принципові особливості функціонуван-
ня і розвитку інтелектуальних систем. Дана порівняльна оцінка деяких з них.
Ключові слова: показники якості, когнітивні технології, інтелектуальна система,
нейронна мережа, вірогідність, інтерполяція, стійкість, структура, швидкодія, надійність.
Introduction
With the development of scientific and technical progress are complicated manufactured
products and technology of industrial production, expanding range, increases the frequency of
turnover of products and technologies, increasing knowledge intensity of production. In indu-
strialized countries, conducted active research in the field of high technologies such as nano-,
bio-, information and cognitive technologies (NBIC). The result is the production of materials
with fundamentally new properties, the development of supercomputers and quantum bio-
computers, intelligent information processing systems. Important from this point of view,
the document is adopted in our country in the 2011 Presidential Decree «On approval of
the priority directions of science and technology in the Russian Federation and the list of
critical technologies of the Russian Federation». One of the priorities recognized «Industry
of nanosystems» and in critical technologies – «nano-, bio-, information, cognitive tech-
nologies».
Standardisation of Indicators of Quality of Products Cognitive Technologies
«Штучний інтелект» 2014 № 3 73
4А
In technical terms, cognitive technologies – technologies that build intelligent systems
(IS), the type of operating the human nervous system and having its capabilities, and in
some respects – superior in the organization of complex behavior in dealing with intel-
lectual problems [1-12]. From a practical point of view, important to establish some
standard range of quality indicators (parameters) adequately characterize the fundamental
features of construction, operation and development of IS, as well as standard methods
(methods) to evaluate these parameters [4], [7], [11]. These issues are currently in the initial
stage of theoretical study, but as market saturation IS their relevance will only increase.
Indicators of quality of IS
The main indicator of properties of IS such as «plasticity», the ability to effectively
solve complex problems and achieve the goal, adapting, altering their behavior as
conditions change, keeping the goal is integrity – law, manifested in appearance, the appearance
of a system of new properties missing from its constituent elements. This figure assumes that
the properties of IS as a certain whole, depend on the properties of its constituent elements,
without being a simple sum of these properties. Furthermore, it is assumed that the
combined elements of the system must have, within certain limits, the variability in the
properties when combined system – they lose part of or acquire new ones. Dual with
respect to the integrity indicator is additive index of the state of the system as it
disintegrated into independent elements. This is an extreme case where the system itself no
longer exists. Real evolutionary system is, as a rule, between the state of absolute integrity
and absolute additivity and generated state of the system (its "slice") can be characterized
by the degree of manifestation of one of these properties, or a tendency to its escalation or
reduction [8]. The evaluation of these trends can be used such indicators as progressive
factorization – aspiration system to a state with an increasingly independent elements, and
progressive systematization – the desire to reduce the independence of the elements, i.e. to
greater integrity. In the system analysis uses comparative quantitative assessment of the
degree of wholeness and of the use factor of the elements properties . Parameter
characterizes the relative connectivity elements of the system and the parameter
characterizes the relative freedom of the elements. In the case of absolute wholeness:
1 , 0 , progressive systematization: , progressing factorization: , absolute
additivity 1,0 . 1 – the basic law of systems.
With indicator of wholeness closely related indicators of IS such as communi-
cativeness and hierarchical pattern. The communicativeness – characterizes the ability of
the system to form a complex unity with the environment. It should be understood that any
system under study is an element of a higher order, and the elements of each system
studied, in turn, generally act as a low-order system. From this it follows that IS is not
isolated from other systems, it is associated with a variety of communications media. A medium
should be understood as the totality of all objects whose properties change affects the system, as
well as those objects whose properties change as a result of the system's behavior.
The hierarchical pattern – characterizes the degree of subordination, structural
ordering system construction. Higher hierarchical level provides guiding influence on the
underlying level, subordinate, and this effect is manifested in the fact that the hierarchy of
subordinate elements acquire new properties that are absent in them in isolation, but as a
result of these new properties, a new, a different aspect of IS. Thus created a new IS acquires the
ability to implement new features, what is the meaning of education hierarchies.
Distinguish a number of important indicators of IS.
The self-organization – characterizes the ability of the system to reach a new level of
development and, in particular, are increasingly manifest properties such as the ability to
Antsyferov S.S.
«Искусственный интеллект» 2014 № 3 74
4А
resist entropy processes and develop anti-entropic (negentropy) trends, adapt to changing
conditions, if necessary, transforming its structure while maintaining certain stability. The
features of self-organization are:
the unsteadiness (variability, instability) of the individual parameters of the system and
its stochastic behavior;
the unique and the unpredictable behavior of the system under specific conditions,
manifested in the system due to the presence of active elements and at the same time
limiting possibilities defined properties of the elements and structural constraints;
the adaptability – the ability to adapt to changing environmental conditions,
interference, as emanating from the environment and influencing the system and internal;
the disequilibrium – the system is fundamentally unstable condition and furthermore,
seeks to maintain in this state;
the negentropic – the ability to resist entropy (deplete system) trends due to the
presence of active elements that promote information exchange with the environment and
the actual process of self-development;
the variability in behavior and structure, the ability to reach a new level of equifi-
nality, while maintaining the integrity and basic properties;
the ability and the willingness to goal formation, when the goals are not set from
the outside, but are formed within the system.
These features are inconsistent, in most cases they are both positive and negative,
desirable and undesirable. They are not immediately possible to understand and explain to
select and create the desired degree of their manifestation. Conflicting particularly
developing systems must constantly monitor and reflect on models and look for methods
and tools that enable them to measure (assess).
The adaptability – the ability of the system to show purposeful, adaptable behavior
in difficult conditions (media). Adapting to the environment allows the system to achieve
this goal in the case of insufficient a priori information about the environment [3, 5, 6]. If
the system can not adapt to changes in the environment, then it dies. In the process of
adaptation may change the quantitative characteristics of the system, its structure, the
functioning and behavior. The most complex form of adaptation have intellectual – self-
organizing systems. In this study of adaptive mechanisms leads to the analysis of complex
problems contradictions stability and freedom of choice, which play an important role in
ensuring the development of the system and adapting to changing conditions, i.e. to study
the stability problem of developing systems [8].
The stability – the ability of the system to return to equilibrium after it was withdrawn
from this state under the influence of external and in systems with active elements –
internal disturbances. The simplest case of steady-state system is a balance – a state in
which the system remains indefinitely in the absence of disturbances. IS are self-organizing
systems with active elements, so their stability should be seen as a reflection of a binary
nature of natural processes: stability – handling, stability – development. In the most
general form the stability of the IS is the ability of system to restore the original or close to
original mode for a small infraction and continue normal operation with a sharp violation
of the regime, keeping its former state qualitatively described system parameters.
The processing speed – is characterized by the following parameters: – time to
return the system to equilibrium after the influence of disturbances; – time to adapt to new
conditions; – time recognition, decision.
The reliability – the ability of the system to maintain its function in case of failure
of any number of constituent elements.
Standardisation of Indicators of Quality of Products Cognitive Technologies
«Штучний інтелект» 2014 № 3 75
4А
Estimation of the properties of IS
The benchmark for comparison when evaluating the properties and functionality of
the IS in cognitive technologies is often the human brain.
The main way of understanding the processes occurring in the brain, understanding
the available experimental data, posing new challenges is the construction and study of
mathematical models that are consistent with the data on the architecture, functions and
features of the brain. Thus, in accordance with nonlinear mathematical models, processes
such as perception, learning, thinking and other brain functions caused by collective
process leading to the coordinated work of ensembles of neurons. Self-organization of
these ensembles and is the key to explaining the functions of the brain. The exceptional
efficiency of the brain in dealing with a huge number of tasks is explained by the work of
the set of "elementary processor" and perfect the system architecture, which effectively
"parallelize" task. Attempts to create computers with a high degree of parallelism –
multiprocessor systems, transputers, cellular automata have shown that the establishment
of parallel algorithms, the use of which part of the computer system do not interfere with
each other and not too long waiting for the results from the other parts, is a very complex
task. Nevertheless, the results of studies in neuroscience are widely used for the creation of
new computer architectures and giving a kind of intuitive computing, associative memory,
ability to learn and generalize the information received [10].
Currently, the most common model in neuroscience is the Hopfield neural network.
Dynamics of Hopfield network is determined by the known relations for discrete
dynamical systems:
,,2,1N,1,
1
1
ti
N
j
itjSijJsigntiS
where 1tiS – the state of the i-th neuron in the time 1t ;
tjS – the state of the j-th neuron in the time t;
ijJ – the link weight of the i-th and j-th neuron;
i – the local threshold at above which the i-th neuron becomes active state;
N – the total number of neurons.
In the Hopfield model, the values i of the threshold parameters are chosen to be
zero, and the matrix relations defined by the rule learning Hebb, when the links are set
primarily between neurons, which when executed by one task are in the same, such as the
excited state. In the Hopfield model, it corresponds to an increase of link weight ijJ , for
example, between the i-th and j-th neuron. According to the model, an important characteristic of
the neural network is the ratio of the number of key images M that can be stored to the number
of neurons of the network N : NM / . For the Hopfield model is often quoted value =
0,14, however, as the experts, actually much smaller. For many nonlinear systems in which
the possibility of collective processes can be distinguished leading variables, the so-called order
parameters, which are adjusted all the other degrees of freedom of the system studied. For the
Hopfield model parameters such as the correlation between the images appear.
The Hopfield model has two major drawbacks. The first is connected with the
existence of «ghosts», «phantoms», «false images», i.e. «false memory». By training the
neural network by Hebb rule, along with the «real», the desired image appears set
Antsyferov S.S.
«Искусственный интеллект» 2014 № 3 76
4А
parasitics. The second shortcoming is that the network is not able to say «i do not know»,
she recognizes, even if it had no reason to. To compensate for these deficiencies is used
generically modified Hopfield model, based on the use of sophisticated neuron model
whose behavior as strong influences (hyperpolarizing or depolarizing) is regular, and at
some intermediate impact – chaotic. Analysis of the generalized Hopfield model showed
that this neural network is able to «fight» with false images, making them unstable corresponding
fixed points. At the same time, the less similar to the presented image previously stored, the
longer it takes for neural network decision. Normally, when the numerical experiments, before
the neural network to recognize the task of the presented randomly in a finite number of steps.
After this time the network may be in one of three states:
correct recognition, the probability of which 1P ;
mistaken recognition whose probability 2P ;
condition «do not know» when the neural network is still in the stage of «reflection»,
whose probability 3P .
The link of these probabilities with the number of steps before a decision is studied in [12].
The main task for the Hopfield model – creating systems with associative memory.
However, there is a wide class of problems, the solution of which it is desirable that the
system had some semblance of intuition and was able to learn from their mistakes. One
such problem is the identification of the function at a given point in space by a set of
known values at other points. Difficulties associated with the fact that the space can have a
very high dimensionality that is typically, for example, for diagnosing problems of
complex disease [1], [2], and this compact space region corresponding to a certain state
(diagnosis) may be sufficiently complex geometry, especially not be convex. Another
problem associated with the forecast value of some quantity for a number of its previously
measured values. Both of these problems can be reduced to the problem of interpolation.
Interpolation theory is one of the most developed areas of computational mathematics.
However, the attempt to use most of the algorithms in the case of high-dimensional space,
or in situations where the numerical values (measurements) are given with an accuracy,
encounters great difficulties [13], [14]. Even the task of finding neighbors in a multidimensional
space may be quite simple and not require the use of special numerical methods.
The neural networks can satisfactorily solve interpolation problems based on them
manage to create recognizing, diagnosing and forecasting systems. . Especially the great
potential of three-layer networks. A key issue for these networks is associated with the
creation of effective learning algorithms. One of the most well-known, simple and reliable
algorithm is backpropagation algorithm, which provides the possibility of correcting the
balance in reverse passage in the case of incorrect recognition. However, this algorithm has
a number of important limitations:
define a local rather than a global minimum of the potential function;
at training multilayer network may be a phenomenon called «network paralysis»,
which manifests itself in the fact that, despite the long time of training, the error may not
decrease substantially, while remaining sufficiently large due to the weak, under certain
conditions, the reaction network correction weights.
The neural networks can be used as a tool for modeling various nonlinear systems.
They are a kind of prototype information complexity [10], typical for a number of physical,
biological and technical systems, simulation model of the process.
Note the common signs of IS of most promising, symbolist-connectivity type. First,
it is a large number of constituent elements and their connectivity, creating new properties
(new quality) of the whole, can not be reduced to the properties of constituents, but at the
same time, in some way determined by the properties of the parts. In turn, the composition
Standardisation of Indicators of Quality of Products Cognitive Technologies
«Штучний інтелект» 2014 № 3 77
4А
and properties of the elements in some way determined by the properties of the whole.
Second, the presence of stable structures and interactions between the elements that
combine elements into a whole. Thus, human brain contains about 1012 neurons can be
represented as a state space 1012 measurements. Typical stable structure consists of
approximately 108 items. In the space of states with 108 measurements and 10 levels of
neuronal excitation exists about
81010 certain provisions, representing excitation vectors.
In the presence of about 105 synaptic connections between each neuron and the other 108
neurons stable structure, it is necessary to distinguish between approximately 1013 synapses.
Consequently, for 10 specific weights at each synapse only one stable structure
there is a huge amount of weights
131010 . This complexity provides limitless opportunities
for encoding, representation and processing of information. Even more of the brain
determined by the number of stable structures (number of connections). According to a
mechanistic approach possible number of links proportional to 2N , where N – number of
elements to be linked, the number of neurons. However, in accordance with the cognitive
approach, the actual number of links NNM ln , i.e. elements "select" the most important
and valuable connections, forming under the influence of the information field connected
together [9], [10].
Thus, the study of American scientists to analyze the visual cortex have revealed
important structural feature in the organization of the brain – the neurons with the same
function are grouped in columns, a sort of connected sets representing local neural
networks, piercing bark. In the theory of cognitive systems is used as an indicator of
quality «cognitive factor» NNNNNNMCF /)(ln2/)ln(2/ , which shows that the
connected set of (stable structure), folding in the information space, give the greater the
effect, the greater the number of elements N they connect. It is assumed that the lower the
value the CF, the greater the effect, the higher the level of intelligence of IS. Evaluating the
level of intelligence as CFLI /1 to the brain 10107,3121027/1 LI .
Assessing the performance, it should be noted that the quantitative characteristics
computer vastly superior brain. The velocity of propagation of the electrical signal is about
810 m/s, and the clock speed of modern computers 910 Hz. The speed of propagation of
the nerve impulse 100 m/s. Nevertheless, as noted above, the work of the brain in solving
intellectual problems much more efficient computer systems.
As for the reliability of IS, there is taken to distinguish the reliability of information
storage and reliability of pattern recognition. The results of studies of human memory and
animals have shown that the brain is no one clearly localized structure for the storage of
information. In this case, the storage system is likely and the distributed nature of the
failure of its components leads to failure of the entire system. Compared with brain
computer components are highly unreliable structures. Failure of any of them may indicate
failure of the entire system. Great prospects improve of indicators of quality of IS, along
with nanotechnology involve using methods biocybernetics.
Abstracts
The set of indicators of quality basic products cognitive technologies – intelligent
systems is proposed.
Comparative assessment of some indicators of artificial intelligence with rates of
natural intelligence is given.
Antsyferov S.S.
«Искусственный интеллект» 2014 № 3 78
4А
Using advances of nanotechnology industry and biocybernetics in IS of neurolo-
gical type will allow, apparently, in the near future much closer to the quality indicators of
natural intelligence.
References
1. Antsyferov S.S. Forming the spectrum of thermal images of objects and recognizing their patterns /
S.S. Antsyferov // Optical jornal. – 1999. – V. 66, № 12. – P. 44-48.
2. Аntsyferov S.S. Forming the spectrum of thermal images of objects and recognizing their patterns /
S.S. Аntsyferov // J. Opt. Technol. – 1999. – V. 66, № 12. – Р. 1047- 1049.
3. Antsyferov S.S. Adaptive information processing technology of thermal broadspectral fields /
S.S. Antsyferov , N.N. Evtikhiev, B.I. Golub // High technology. – 2002. – V. 3, № 4. – P. 45-50.
4. Antsyferov S.S. Metrology of virtual systems // Measuring equipment. 2003. № 5. P. 17-21.
5. Antsyferov S.S. System quality management principles of designing adaptive information-recognition systems /
S.S. Antsyferov, A.S. Sigov, E.S. Antsyferov, B.I. Golub // Proceedings TRTU. – 2005. – № 10. – P. 167-174.
6. Antsyferov S.S. Adaptive information processing of spatiotemporal isotropic fields / S.S. Antsyferov,
N.N. Evtikhiev // Optical jornal. – 2006. – V. 3, № 10. – P. 52-57.
7. Antsyferov S.S. Metrology of intellectual systems / S.S. Antsyferov // Artificial Intelligence. – 2008. –
№ 3. – P. 18-27.
8. Antsyferov S.S. General construction principles and regularities of functioning of intellectual systems /
S.S. Antsyferov // Artificial Intelligence. – 2011. – № 3. – P. 6-15.
9. Antsyferov S.S. Questions of metrological provision of intellectual systems / S.S. Antsyferov // World of
measurements. – 2012. – № 5. – P. 46-51.
10. Antsyferov S.S. Metrological basis of forming of information space in intellectual processing systems /
S.S. Antsyferov // Artificial Intelligence. – 2012. – № 4. – P. 283-291.
11. Antsyferov S.S. Estimation of quality level of intellectual systems / S.S. Antsyferov // Artificial Intelligence. –
2013. – № 3. – P. 316-323.
12. Antsyferov S.S. Processing speed of intellectual systems / S.S. Antsyferov, K.E. Rusanov // Artificial
Intelligence. – 2013. – № 4. – P. 259-265.
13. Antsyferov S.S. The general theory of measurement / S.S. Antsyferov, B.I. Golub. – M. : Hotline-
Telecom. 2007. – 176 p.
14. Antsyferov S.S. Fundamentals of theoretical metrology / Antsyferov S.S., Afanasiev M.S. – M. :
«Icarus», 2012. – 208 p.
RESUME
S.S. Antsyferov
Standardisation of Indicators of Quality
of Products Cognitive Technologies
In technical terms cognitive technologies is the technology of construction of
intellectual systems (IS), functioning according to the type of the human nervous system
and possessing its guests, and in some respects – exceeding in the organization of complex
behavior in solving intellectual problems. From a practical point of view seems to be
important to establish some standard nomenclature of indicators of quality (parameters),
adequately characterizing the principal features of the building, functioning and
development IS, as well as standard methods (techniques) assessment of these indicators.
These questions are currently in the initial stage of theoretical development, but as the
saturation of the market IS their importance will only grow.
The basic indicators of the quality of IS may act integrity, hierarchy, self-organization,
adaptability, stability, processing speed and reliability. The standard of comparison in evaluating
the performance of the quality of IS serves as the brain of the person or its mathematical model.
The most common model is the Hopfield neural network.
Standardisation of Indicators of Quality of Products Cognitive Technologies
«Штучний інтелект» 2014 № 3 79
4А
The main task for Hopfield model is the creation of systems possessing associative
memory. However, there is a wide class of problems, for solution of which it is desirable
that the system had a certain similarity of intuition and was able to learn from their
mistakes. One of such tasks associated with the definition of the function values at a given
point in space at a known set of values at other points. Difficulties related to the fact that
the space can be very high dimension, which is typical for the problems of diagnostics of
complex diseases, while a compact region this space corresponding to certain states
(diagnoses), can be rather complicated geometry, in particular, not be convex. Another
challenge is to forecast the values of some size on a number of its previously measured
values. Both of these problems are reduced to the problem of interpolation.
The neural networks are quite satisfactory to solve interpolation tasks, on their basis
it is possible to create recognize, diagnostic and prediction systems. Especially the great
potential of three-layer networks. The main question for these networks is associated with
the creation of effective learning algorithms.
The neural networks can be used as a tool for the simulation of different nonlinear
systems. They are a kind of prototype of informational complexity typical for a number of
physical, biological and technical systems simulation model of the investigated process.
Use of the achievements of nanotechnology and bio-cybernetics in IS of
neurological type will, apparently, in the nearest future much closer to indicators of quality
of natural intellect.
The article entered release 09.04.2014.
|