Abstracts
Збережено в:
| Дата: | 2013 |
|---|---|
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України
2013
|
| Назва видання: | Кибернетика и вычислительная техника |
| Онлайн доступ: | http://dspace.nbuv.gov.ua/handle/123456789/84487 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Abstracts // Кибернетика и вычислительная техника. — 2013. — Вип. 173. — С. 93-97. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
irk-123456789-84487 |
|---|---|
| record_format |
dspace |
| spelling |
irk-123456789-844872015-07-09T03:02:03Z Abstracts 2013 Article Abstracts // Кибернетика и вычислительная техника. — 2013. — Вип. 173. — С. 93-97. — англ. 0452-9910 http://dspace.nbuv.gov.ua/handle/123456789/84487 en Кибернетика и вычислительная техника Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| format |
Article |
| title |
Abstracts |
| spellingShingle |
Abstracts Кибернетика и вычислительная техника |
| title_short |
Abstracts |
| title_full |
Abstracts |
| title_fullStr |
Abstracts |
| title_full_unstemmed |
Abstracts |
| title_sort |
abstracts |
| publisher |
Міжнародний науково-навчальний центр інформаційних технологій і систем НАН України та МОН України |
| publishDate |
2013 |
| url |
http://dspace.nbuv.gov.ua/handle/123456789/84487 |
| citation_txt |
Abstracts // Кибернетика и вычислительная техника. — 2013. — Вип. 173. — С. 93-97. — англ. |
| series |
Кибернетика и вычислительная техника |
| first_indexed |
2025-07-06T11:28:58Z |
| last_indexed |
2025-07-06T11:28:58Z |
| _version_ |
1836896841251160064 |
| fulltext |
93
ABSTRACTS
DEDICATED TO THE 90TH BIRTH ANNIVERSARY
OF ACADEMICIAN V.M. GLUSHKOV
On 24 August 2013 ninety years passed since Victor Glushkov’s birthday, who
was an outstanding scientist, author of fundamental works in the field of cybernetics,
computer engineering and applied mathematics. Academician V.M. Glushkov lived to
be 58, and 35 years of his creative development brought hundreds of works which
contributed to forming cybernetics as a science.
V.M. Glushkov joined such important courses of scientific-technical progress as
the cybernetics fundamentals development and practical methods of developing
computer engineering forming. He was one of the founders of the interdepartmental
collection of scientific papers “Cybernetics and Computer Engineering”. Thoughtful
scientist and unsurpassed pedagogue V.M. Glushkov is a founder of a world-known
school.
UNESCO ANNOUNCED 2013 AS THE YEAR OF N.M. AMOSOV
IN MEDICINE
On 6 December 2013 we will celebrate Nikolai Amosov’s 100th birthday
anniversary who was an outstanding scientist and genius surgeon, founder of
cardiosurgery in Ukraine, medical and biological cybernetics, writer and public figure.
The interdepartmental collection of scientific papers “Cybernetics and Computer
Engineering” is presenting a series of articles devoted to the development of scientific
schools initiated by N.M. Amosov. The basic results of the articles were discussed and
recommended to be published at the seminar “Biological and Medical Informatics and
Cybernetics” (18-21 July 2013, Zhukin) dedicated to the anniversary.
V.I. Gritsenko, D.A. Rachkovskij, A.D. Goltsev, V.V. Lukovych,
I.S. Misuno, E.G. Revunova, S.V. Slipchenko, A.M. Sokolov
NEURAL DISTRIBUTED REPRESENTATION FOR INTELLIGENT
INFORMATION TECHNOLOGIES AND MODELING OF THINKING
Introduction: The state-of-the-art arsenal of artificial intelligence and machine
learning includes well developed methods for vector data. However, traditional vector
representations of high-dimensionsioanl data lead to slow. On the other hand,
processing of complex structured data represented in a symbolic form requires
sequential and often computationally complex algorithms (e.g., graph similarity is often
determined by the sub-graph isomorphism, which is
NP-complete problem). So, such methods hardly applicable to large-scale applications.
The purpose of the paper is to review our approach to creation of intelligent
information technologies based on modeling human thinking and a special kind of
vector representations that use the idea of distributed representations in the brain. This
approach is the development of the ideas of Nikolai Mikhailovich Amosov and his
scientific school.
Results: It is shown that distributed representations allow an efficient estimation
of numeric vector similarity as well as efficient solving of discrete
ill-posed problems. The formation of distributed representations of complex structured
relational information used in declarative knowledge bases is considered, that allows an
efficient search for analogs in analogical reasoning.
Possible implementation of N.M. Amosov's functional acts for modeling of
intelligent behavior based on distributed representations is also proposed.
94
Conclusions: Thus, distributed representations, when used in information
technologies, can improve computational efficiency by converting various data types
(both unstructured information in the form of vectors arrays and relational structures of
knowledge bases) to a special format of binary vectors. In addition, distributed
representations can naturally combine information about the structure and semantics,
opening the possibility of creating computationally efficient and qualitatively new
methods for processing relational structures of data and knowledge bases by similarity
of their representations. Neurobiological relevance of distributed representations opens
the possibility of using them as the basis of intelligent information technologies that
function similarly to the human brain.
Keywords: distributed representations, similarity-based, retrieval, analogical
reasoning, discrete ill-posed problem.
A.D. Goltsev, V.I. Gritsenko
ALGORITHM OF SEQUENTIAL FINDING THE TEXTURAL
FEATURES CHARACTERIZING HOMOGENEOUS TEXTURE SEGMENTS
FOR THE IMAGE SEGMENTATION TASK
Introduction: The paper describes a part of the model for texture segmentation of
visual images. The model is designed to solve the problem of segmentation of arbitrary
images into homogeneous texture segments under condition of absence of any
information about the image. A computer program simulating the texture segmentation
model is created. The program has been tested on black and white natural images of a
landscape type. The experiments demonstrate the effectiveness of the model for texture
segmentation of natural images.
Purpose: The texture segmentation procedure is performed basing on textural
features extracted from the image. The segmentation process is sequential and iterative,
one homogeneous texture segment is delineated in each iteration. At the beginning of
every iteration, a set of textural features characterizing the currently extracted segment
is found by means of some computations in a multi-dimensional feature space. Thus, it
is development of the algorithm for finding a representative feature set that is the goal
of this work.
Results: The algorithm for finding a representative feature set that characterizes
homogeneous texture segments has been designed; this algorithm is a part of the model
for segmentation of arbitrary image into homogeneous texture regions under condition
of absence of any information about the image.
Conclusions: The proposed algorithm is a key part of the texture segmentation
model. The algorithm provides automatic finding the texture features that characterize
homogeneous texture segments presenting in the analyzed image. It is the algorithm
due to which the model can do without specifying the number of segments to which the
image should be divided.
Keywords: texture; texture window, textural feature, texture segmentation, pixel.
E.G. Revunova
RANDOMIZATION APPROACH TO THE RECOVERY OF SIGNALS
RESULTED FROM INDIRECT MEASUREMENTS
Introduction: A number of applications require solving the problem Ах ≈ у,
where the matrix A and the vector y (distorted by additive noise) are known. We need
to estimate the signal vector x. For example, it may be a linear measuring system with
the measurement result y. The matrix A of the input-output linear transformation
describes the interaction of the measured signal with the medium and measuring tool.
The columns of A can be viewed as discrete samples of the basis functions of the
measurement system. Another problem appears when we know desired set of basis
95
functions C that would provide improved resolution of the measuring system. The task
is to find out the observed output (obtained with A) to the output of the system with the
given basis C.
For some special A observed in practice the solution based on matrix inverse is
unstable, and regularization is needed. The classic method is Tikhonov regularization.
However it suffers from a high computational complexity and the difficulties in
selecting the proper regularization parameter. So, the purpose is to develop approaches
to regularization that provide solution accuracy at the level of Tikhonov regularization,
but with less computational complexity.
Results: This paper proposes an approach to stable solution of the output
transformation problem based on the use of random projections and
QR-decomposition. An experimental study of the solution error, its components, and its
dependence on the dimension of the projector matrix is conducted. For a randomized
method of solving discrete ill-posed problem, the expressions for error of output signal
recovery are obtained, and an experimental study of the error is also conducted.
Conclusions: The proposed approach based on a distributed neural network
representations and random projections enables the development of efficient
regularization methods both for solving the problem of recovering the signals resulted
from indirect measurements and the problem of transforming the observed outputs of a
measurement system.
Keywords: projector, random matrix, regularization.
E.M. Kussul, T.N. Baidyk
MICROMECHANICS AS A TESTBED FOR EVALUATION OF
ARTIFICIAL INTELLIGENCE METHODS IN MANUFACTURING
Introduction: Artificial intelligence (AI) methods can be used to improve
automation systems in manufacturing processes. However, the application of these
methods in industry is not widespread, because of the high cost of experiments with AI
systems applied to conventional manufacturing systems. To reduce the cost of
experiments in this area, we have developed a specific micromechanical equipment,
similar to conventional mechanical equipment, but of much smaller size and therefore
of lower cost. This equipment can be used for evaluation of different AI methods in an
easy and inexpensive way. The methods that show good results can be transferred to
the industry through appropriate scaling.
The purpose is to provide a brief description of low cost microequipment
prototypes and some AI methods that can be evaluated with such prototypes.
Results: Several neural network algorithms were proposed to improve automation
systems in manufacturing processes. These algorithms were tested with specific
micromechanical equipment, similar to conventional mechanical equipment, but of
much smaller sizes and therefore of lower cost.
Conclusions: We consider this equipment a good testbed for examination of the
AI algorithms that can be used to increase the level of automation of manufacturing
processes. One of the problems we intend to examine is the prediction of resonant
oscillations in the process of turning and avoidance of resonance vibrations using
assembly neural networks.
Keywords: artificial intelligence, micromechanics, computer vision,
neural networks, neural assemblies.
S.V. Slipchenko
NAMED ENTITY RECOGNITION USING CONTEXT VECTORS
Introduction: Named Entity Recognition is one of the main tasks of natural
language processing. Most of the statistical methods are used to detect a huge number
96
of local features (millions and tens of millions), and rely heavily on secondary
characters (mixed case characters, special characters, etc.).
The purpose is to investigate the quality of named entity recognition using
Conditional Random Fields with local context features and develop a new approach to
taking into account global context features using context vectors.
Results: Our experiments show that recognition precision by Conditional Random
Fields is good enough for word only local features. To take global context into account,
we propose using context vector of the words represented by localist and distributed
vector representations.
Conclusions: The proposed approach that include various combinations of localist
and distributed representation of local and global context, as well as informative feature
selection, should be investigated to improve the efficiency of learning.
Keywords: distributed representations, named entity recognition.
I.I. Yermakova, J.P. Tadejeva
MODELING COMPLEX FOR PREDICTION HUMAN UNDER EFFECTS
OF GENERAL AND REGIONAL ULTRASOUND-FREQUENCY
ELECTROMAGNETIC FIELD
Introduction: Ultrasound electromagnetic radiation is commonly used in
medicine. Evaluating human exposure to ultrasound monitoring is needed in order to
achieve the required temperature in irradiated array at a certain depth, as well as
evaluation of the exposure human endurance.
The purpose is to study the result of the general and regional human exposure to
ultrasound with the developed mathematical models.
Results: A method for modeling specific absorption rate of an ultrasonic energy
by tissues is developed. Mathematical models of energy and heat processes of human
for the effect of ultrasonic treatment taking account the characteristics of
electromagnetic field in ultrasound ranges are performed.
The developed modeling complex taking into account the responses of nervous
system, cardiovascular system, water-salt exchange system, thermoregulatory system
and characteristics of ultrasonic exposure makes prognosis of preliminary physiological
status and allows to evaluate the result of hyperthermia treatment.
Conclusion: Analysis of the results of computational experiments with the results
of other researchers showed that the developed 34-compartments mathematical model
for prediction of general and regional human exposure to ultrasound adequately
describes the effect of ultrasonic radiation on humans.
Keywords: modeling complex, ultrasound frequency, electromagnetic
hyperthermia, specific absorption rate, whole body hyperthermia, regional
hyperthermia, prediction of thermophysiological human state, mathematical model.
V.V. Lukovych
SIMPLE ARCHITECTURE OF CONVOLUTION NEURAL NETWORK
FOR HANDWRITTEN DIGIT RECOGNITION
Introduction: Convolutional neural network is an example of architecture that is
specialized for image recognition. Universal classifier is often used in the last stages of
convolutional neural networks. The last feature map of a convolutional neural network
contains data whose dimension is usually high. Therefore, it can be assumed that a
simple linear classifier could work effectively there.
Purpose: We proposed a simple convolutional neural network with 1-layer
classifier in the last stage. The aim of this work was to investigate the performance of
such neural network.
97
Results: A series of experiments was carried out to evaluate proposed network on
the MNIST database. An error rate of about 0.38% was achieved.
Conclusions: Experiments have shown that proposed network can yield
performance comparable to the state-of-the-art of handwritten digit recognition.
Keywords: convolutional neural network, handwritten digit recognition, MNIST
database.
K.G. Lyabakh
MATHEMATICAL MODELS FOR INVESTIGATION OF INFLUENCE
OF NITRIC OXIDE ON THE OXYGEN REGIME OF CELL
Introduction: The nitric oxide (NO) plays the key role in many biological
functions. NO inhibits mitochondrial O2 consumption (VO2) and may act as a
citoprotector but it may disturb tissue energy supply in dependence on its
concentration[NO]. The existence of NO in tissue is connected with O2 presence.
Scavenging and NO production in tissues by myoglobin Mb vary with the O2
concentrations and influence oxygen regime parameters: pO2, NO and VO2 distribution
in tissue.
The purpose of our study was to develop the model of O2 and NO transport and
interaction between NO and Mb in the muscle cell for calculation of main
characteristics of oxygen regime in the cell — O2 parameters.
Results and discussion: Two computer models of O2 and NO delivery to and O2 -
consumption and NO- elimination in the working muscle cell have been developed:
The model 1 describes convective O2 delivery by blood and three-dimensional O2
and NO diffusion-reactions and in the muscle fiber with reversible inhibition of
mitochondrial respiration by NO without Mb. Using this model we obtained [NO] and
pO2 values away from capillary in the muscle fiber, taking into consideration the
mutual influences [NO] and [O2 ] on each other under the parametric changes of [NO]
from 0 to 60 nM. We investigated an inhibitory effects of NO on tissue respiration.
Model 2 described the influence of myoglobin on [NO] and [O2] in the cell. Planar
radial oxygen, NO and myoglobin diffusion were described. Nitrite reductase and
oxidase functions of myoglobine are taken into concideration. Acting as an NO
scavenger in normoxia condition, oxygeneted myoglobin locally decreased [NO] and
eliminated NO-inhibition of mitochondrial activity. Under hypoxia deoxygenated
myoglobin produced NO only in the zone of oxygen lack.
Conclusions: According to calculations intracellular NO gross impacts the oxygen
regime of cell, because it increases tissue pO2 and depresses mitochondrial respiration
in the whole cellular volume. Myoglobin modulates the influence of NO on oxygen
regime and provides a fine adjustment of oxygen regime in muscle, increasing tissue
pO2 and decreasing a mismatch between oxygen supply and demand.
Keywords: Mathematical model, oxygen mode, oxygen transport, nitric oxide,
myoglobin.
|