Wyniki wyszukiwania - Biblioteka Nauki

1

On the accuracy function and the accuracy radius for combinatorial optymization problems

100%

Libura M.

|

2000

|

tom Vol. 48, nr 2

231-245

EN

We consider the combinatorial optimization problem in the following form: min{EEF c(e) : F E F} -( formula), where for some finite set E, c = (c(e), e E E) E R+ (formula) is a vector of weights of elements of E, and F C 2E (formula) is a given family of feasible subsets. We assume that the set F is fixed, but the weights of elements belonging to some given subset X C E may vary. The accuracy function describes the quality of a given solution of the problem as a function of weights perturbations. Namely, for a given optimal solution Fo obtained for an initial vector of weights ca E R+, the value a( Fo, X, b) of this function for some specified subset X C E and a value b, where 0 < 1, is equal to the maximum relative error of the solution Fo for arbitrary simultaneous perturbations of weights of elements, satisfying the following conditions: Ic(e) - co(e)I

2

Modelling of nanomaterials - sensitivity analysis with respect to the material parameters

80%

Kopernik M. , Szeliga D.

|

2007

|

tom Vol. 7, No. 2

255-261

EN

Nanomaterials are challenging in modelling and in experiments. Because of many difficulties in measuring their properties, there is a necessity to use computing methods to obtain better results. Sensitivity analysis, as a numerical method, will be used to find material model of nanocoating and to do the optimization of deformation process.

PL

Badania doświadczalne nanomateriałów są wyzwaniem dla naukowców. Ze względu na trudności związane z interpretacją wyników pomiarów, na podstawie których wyznaczane są własności nanomateriałów, i ich duży rozrzut, wykonywane są komputerowe symulacje doświadczenia w celu dokładnej analizy zachodzących zjawisk. W niniejszej pracy analizowano cienkie powłoki dla modelowania instrumentalnej próby wciskania wgłębnika i przewidywania zależności siły od głębokości wciskania. Doświadczenie symulowano metodą elementów skończonych, przyjmując zaokrągloną końcówkę wgłębnika i stosując aktualizację siatki (remeshing). Do optymalizacji próby używana jest zależność siły od głębokości wciskania. W pracy przeprowadzono analizę wrażliwości siły względem parametrów procesu dla identyfikacji tych parametrów, które mają istotny wpływ na przebieg doświadczenia. W wyniku analizy określono wpływ parametrów sprężystych i kształtu wgłębnika na wyniki symulacji.

3

Shape optimization for dynamic contact problems

80%

Myśliński A.

Discussiones Mathematicae, Differential Inclusions, Control and Optimization

|

2000

|

tom 20

|

nr 1

79-91

EN

The paper deals with shape optimization of dynamic contact problem with Coulomb friction for viscoelastic bodies. The mass nonpenetrability condition is formulated in velocities. The friction coefficient is assumed to be bounded. Using material derivative method as well as the results concerning the regularity of solution to dynamic variational inequality the directional derivative of the cost functional is calculated and the necessary optimality condition is formulated.

4

Frequency optimization based on “exact” sensitivity and fatigue life estimation

80%

Bednarek T. , Sosnowski W.

|

2007

|

tom Vol. 7, No. 2

243-249

EN

In this paper exact sensitivity and optimization together with fatigue calculations of vibrating structures are considered. The finite element formulation of natural frequencies, its sensitivity and application are presented. It is known that natural frequencies of free vibrations can be treated as eigenvalues of some generalized eigenproblem [1,2]. Reduction algorithms of the eigensystem and eigenvalues sensitivity are used. Sensitivities are calculated “exactly” by Direct Differentiation Method (DDM). Results are compared with theoretical values and with other approximate methods of sensitivity calculation (Finite Difference Method and semi–analytical method)[2]. DDM is implemented by the first author into the finite element program FEAP. Optimization algorithm based on sensitivity analysis is also included in the form of external code [3]. A clamped beam, stiffened plate and double – disk rotor shaft system are taken as examples. The objective of optimization is to avoid resonance which may be caused by external varying load. Resonance can be followed by big stress fluctuations in structure, which result in shortening of fatigue life and leads to collapse of the structure. Positive and negative secondary effects of optimization are considered. Dynamic fatigue analysis of the structures before and after optimization are done. Some idea related to equivalent fatigue life of materials and structures loaded by different stress ratio R and counting rainflow algorithm are also implemented into the FEAP code [4]. Frequency optimization can be useful in design process to assist to the engineer. Optimization with objective to avoid resonance can eliminate undesirable noise, vibrations etc. Is specific structures this kind of optimization produce additional positive effect: reduction of the weight of the system with simultaneous decreasing of stress level. Also some possible negative influence of optimization on consecutive eigenpairs (not taken previously into account during optimization) are considered. Resonance elimination may increase fatigue life of the structure. [1] T. D. Hien, Deterministic and Stochastic sensitivity in computational structural mechanics, IFTR Reports 46/1990, Warsaw 1990. [2] M. Kleiber, H. Antunez, T. D. Hien, P. Kowalczyk, Parameter sensitivity in nonlinear mechanics, John Wiley & Sons, Chichester, 1997 [3] J. S. Arora, Introduction to optimum design, McGraw Hill Book Company, 1988 [4] I. Marczewska, T. Bednarek, A. Marczewski, W. Sosnowski, H. Jakubczak and J. Rojek, Practical fatigue analysis of hydraulic cylinders and some design recommendation, Int. Journal of Fatigue (in press, available online).

PL

Przedmiotem artykułu jest optymalizacja ze względu na częstotliwości drgań własnych dowolnych konstrukcji z wykorzystaniem ścisłych metod analizy wrażliwości. Zbadany został wpływ optymalizacji na wytrzymałość zmęczeniową. Celem optymalizacji było uniknięcie rezonansu, który może prowadzić do wzrostu naprężeń w konstrukcji i przedwczesnego zniszczenia zmęczeniowego. Wrażliwość wartości własnych ze względu na parametry projektowe została obliczona Metodą Bezpośredniego Różniczkowania (Direct Differentiation Method, DDM) wcześniej rozwijaną przez Sergeyeva i Mroza, (2000), Kleibera i innych (1997) oraz Sosnowskiego i Kleibera (1996). Jako przykład wybrany został wał wirnikowy z dwoma dyskami sztywnymi, podparty na dwóch łożyskach ślizgowych. Parametrami projektowymi były średnice poszczególnych sekcji wału. Proces optymalizacji miał na celu oddalenie częstotliwości drgań własnych wału tak daleko jak to możliwe (przy zachowaniu odpowiednich ograniczeń) od założonej częstotliwości wymuszenia. Następnie optymalny wał został obciążony siłą zmienną w czasie. Zmiany naprężenia w czasie zostały zapisane, cykle naprężeń zostały zliczone algorytmem "rainflow". Następnie dokonano porównawczej analizy zmęczeniowej wału przed i po optymalizacji.

5

Critical evaluation and sensitivity analysis of rheological models of human blood

80%

Szeliga D. , Kopernik M. , Pietrzyk M.

|

2009

|

tom Vol. 9, No. 4

435-450

EN

Critical analysis and classification of rheological models of a human blood is the objective of the paper. In the first part of the paper main features of a blood and their influence on modelling blood flow are discussed. Various models available in the scientific literature have been analysed and classified on the basis of a mathematical form of equations. Capabilities of the models to account for certain physical features of blood were evaluated. Power laws, commonly used in basic simplified simulations, have simple mathematical form and are not analysed. Among the remaining models, three groups were distinguished and selected for further analysis: Casson type models, be-exponential law and Quemada model. Sensitivity analysis was performed for these models using Morris OAT Design method. Sensitivity of the viscosity predicted by various models, with respect to the coefficients in these models and with respect to the external variable (shear rate), was determined. Importance of the investigated coefficients and their influence on models’ predictions was evaluated. Suggestions concerning identification of coefficients in the models are given in conclusions.

PL

Celem pracy jest krytyczna analiza oraz klasyfikacja reologicznych modeli krwi. W pierwszej części artykułu omówiono główne cechy i właściwości krwi w aspekcie ich wpływu na modelowanie przepływu krwi. Analizie poddano różne modele krwi dostępne w literaturze, które sklasyfikowano w oparciu o matematyczną postać zastosowanych równań. Oceniono możliwości poszczególnych modeli w zakresie uwzględnienia wpływu fizycznych cech krwi. Prawa potęgowe, powszechnie stosowane w symulacjach przepływu krwi, maj ą prosty matematyczny zapis i nie były szczegółowo analizowane. Do dalszej analizy modele podzielono na trzy grupy: modele typu Cassona, prawo Bi-ekspotencjalne i model Quemady. Dla wybranych modeli przeprowadzono analizę wrażliwości stosując metodę opartą o algorytm Morrisa. Określono współczynniki wrażliwości lepkości krwi względem współczynników modeli oraz względem zmiennej zewnętrznej prędkości ścinania. Wyznaczono wskaźniki istotności poszczególnych współczynników oraz ich wpływ na wyniki modelu. We wnioskach z pracy zawarto wskazówki dla identyfikacji współczynników analizowanych modeli.

6

Parameter sensitivity formulation for viscoelastic structures

80%

Tauzowski P. , Kleiber M.

|

1999

|

tom Vol. 6, No. 2

189-203

EN

The problem of sensitivity of viscoelastic response with respect to material parameters is studied in the paper. The direct differentiation method is employed. The FEM-related implementation issues are discussed. A number of numerical examples illustrates the theory.

7

Sensitivity to correlation in multivariate models

80%

Žanic V. , Žiha K.

|

1998

|

tom Vol. 5, nr 1

75-84

EN

The paper presents some aspects of the sensitivity analysis within the multivariate distribution models. The presented procedures are provided for engineering problems based on the Nataf model. The Nataf model involves the marginal distributions of the random variables and the correlation between them. Sensitivities are considered through derivatives with respect to the correlation coeffcients. The terms for the derivatives of the Nataf correlation coefficiens with respect to the given correlation coefficients are presented. The derivatives of the transformations between the random variables are given nex. The Cholesky decomposition and the spectral decomposition are applied. Derivatives of the Cholesky decomposition are obtained in the form of a recursive scheme. Dervatives of the eigenvalues and eigenvectors are obtained using perturbations. In addition, a comprehensive method for derivatives of distances and derivatives of angles between the directions is given. Finally, numerical examples are attached to illustrate the presented procedures.

8

Application of sensitivity analysis to the identification of the Neumann condition

80%

Majchrzak E. , Freus K.

|

tom z. 72

201-204

9

Sensitivity Analysis and Optimization for Selected Supply Chain Management Issues in the Company – Using System Dynamics and Vensim

80%

Kasperska E. , Mateja-Losa E. , Marjasz R.

Journal of Entrepreneurship, Management and Innovation

|

2013

|

tom 9

|

nr 2

29-44

EN

The aim of our paper is to present the new results of research work on optimization and simulation for some logistic problems in the company. The System Dynamics (SD) method and the Vensim simulation language are applied in order to solve specific managerial problems described by Forrester in the model of supply chain. The historical model of Customer-Producer-Employment System by Forrester (Forrester, 1961) has not been examined with the sensitivity analysis, from the “automatic” testing perspective. Optimization experiments have not been conducted, either. It is surprising, since the model is old and widely known. The opportunities offered by the Vensim language allow us to perform such analysis. The visualization called “confidence bounds“ is used, to show the behaviour of chosen variables over a period of time. The Monte-Carlo method is applied for sampling a set of numbers from within bounded domains (distribution for each searching parameters is specified). The authors of this paper conducted numerous experiments in this scope. This paper presents their results and offers some conclusions formulated at the end.

10

Sensitwity analysis in a fatigue delamination problem of an elastic two-layer composite

80%

Figiel Ł. , Lauke B. , Kamiński M.

|

tom Vol. 56, iss. 1

65-87

EN

Several parameters can affect the fatigue delamination growth in laminates - these in-clude e.g. constituent material properties and/or composite shape. Knowledge about effects of these parameters can lead to a better understanding of the fatigue delamination behaviour and can also pinpoint directions for optimum composite design. These effects can be elucidated by carrying out an appropriate sensithdty analysis. A FEM-based computational approach to sensitivity analysis is proposed in this work to study composite parameter effects in a fatigue delamination problem of an elastic two-layer composite. It is used to calculate and analyse sen-sitivity gradients of the fracture parameter and fatigue cycle number with respect to composite design parameters such as layer elastic constants. It is observed that sensitivities computed from this approach are generally numerically stable. Obtained sensitivities pinpoint ąuantita-tively the most and least important composite parameters that govern a fatigue delamination process. Sensitivity results are verified by another computational approach and a very good agreement is found.

11

Selection of the appropriate phase transformation model for design of laminar cooling and continuous annealing of DP steels

80%

Szeliga D.

|

2012

|

tom Vol. 12, No. 2

70-84

EN

In the work modified JMAK phase transformation model dedicated to dual phase steels was proposed. The model was validated with sensitivity analysis methods and the most important model parameters were selected. The sensitivity analysis results were next used in the parameters identification task. That task was formulated as an inverse problem, it was transformed to the optimization task. The minimum of the goal function defined in the optimization problem was the solution of the model parameters identification. The phase transformation model outputs with identified parameters were compared to measurements from dilatometer and microsections analysis. The comparison was satisfied. Two different case studies were defined to apply modified JMAK model to finite element software and to performed simulations of selected case studies. Investigated procedure proved the fitness for modelling of complex industrial processes combining material forming and cooling/annealing.

PL

W pracy przedstawiono zmodyfikowany model przemian fazowych oparty na równaniu JMAK, umożliwiający modelowanie procesu laminarnego chłodzenia i wyżarzania stali dwufazowych. Dla modelu przemian wyznaczono jego najistotniejsze parametry korzystając z metod analizy wrażliwości. Następnie wyznaczono parametry modelu przemian dla stali dwufazowej. Zagadnienie identyfikacji parametrów modelu zdefiniowano jako zadanie odwrotne korzystając z wcześniejszych wyników analizy wrażliwości. Ilościowe oszacowanie parametrów zostało zweryfikowane na podstawie pomiarów dylatometrycznych oraz analizy obrazów mikrostruktury. W drugiej części pracy zaprezentowano obliczenia dla dwóch różnych procesów przemysłowych z wykorzystaniem modelu przemian fazowych połączonym z metodą elementów skończonych. Otrzymane wyniki potwierdziły przydatność opracowanych modeli do modelowania złożonych procesów przemysłowych łączących formowanie materiału z procesem chłodzenia bądź wyżarzania.

12

Sensitivity analysis of biological tissue freezing process with respect to the radius of spherical cryoprobe

80%

Majchrzak E. , Kałuża G.

|

2006

|

tom Vol. 44 nr 2

381-392

EN

Application of the shape sensitivity analysis to the case of problems of the biological tissue freezing process is discussed. The freezing process described by a strongly non-linear bioheat transfer equation in which an additional term controlling the evolution of latent heat appears. Using the approach called 'the one domain method', one finally obtains a partial differential equation containing the substitute thermal capacity of tissue. The boundary and initial conditions determine the thermal interaction between the tissue and cryoprobe tip. In the paper, we consider a spherical internal cryoprobe. In order to estimate the influence of cryoprobe geometry on the course of the process, the shape sensitivity analysis is applied. In particular, a direct approach is used (explicit differentiation method). The results of numerical modeling (the boundary element method is applied) allow one to formulate essential practical conclusions concerning the course of cryosurgery treatments.

PL

Praca dotyczy zastosowania analizy wrażliwości kształtu w zagadnieniach modelowaniaprocesuzamrażania tkanki biologicznej. Proces ten jest opisany silnie nieliniowym równaniem przepływu biociepła, w którym pojawia się dodatkowy składnik związany z wydzielaniem się utajonego ciepła zamrażania. Formalne przekształcenia wyjściowego równania opisującego proces prowadzi do równania różniczkowego, w którym występuje zastępcza pojemność cieplna tkanki (tzw. metoda jednego obszaru). Odpowiednio przyjęte warunki brzegowe determinują oddziaływania termiczne między tkanką a końcówką kriosondy. W pracy rozważa się sferyczną kriosondę wewnętrzną. VV celu oszacowania wpływu geometrii kriosondy na przebieg procesu zamrażania wykorzystano podejście bezpośrednie analizy wrażliwości kształtu. Wyniki obliczeii numerycznych otrzymane za pomocą metody elementów brzegowych pozwoliły na sformułowanie wniosków przydatnych w praktyce kriochirurgicznej.

13

Problemy analizy wrażliwości konstrukcji cienkościennych

80%

Szymczak C. , Mikulski T. , Iwicki P.

|

2000

|

tom z. 14

129-134

PL

Przedstawiono przegląd zastosowań analizy wrażliwości pierwszego rzędu do konstrukcji cienkościennych o przekroju otwartym. Przyjęto podstawowe założenia klasycznej teorii prętów cienkościennych o przekroju nieodkształcalnym. Podano przykłady wyznaczania wariacji przemieszczeń, sił wewnętrznych, obciążeń krytycznych oraz częstości i postaci drgań skrętnych prętów cienkościennych od zmian rożnego typu zmiennych projektowych zarówno w ujęciu ciągłym jak i dyskretnym. Przedyskutowano dokładność otrzymanych rozwiązań oraz możliwość ich uogólnienia na ramy i ruszty.

EN

A review of application of first order sensitivity analysis for thin-walled structures with open cross-section is presented. The classical assumptions of theory of thin-walled bars with nondeformable cross-section are taken into consideration. Some examples of determination of variations of displacements, internal forces, critical loads, frequencies and modes of torsional vibration of thin-walled due to different type of design variables, both continuous and discrete, are given. An accuracy of the solutions obtained and possibility of its generalization on frames and grid structures are discussed.

14

The application of the Morris method for determination of ranges of parameters variation, having assumed influence of calculation result

80%

Sczygiol N. , Dyja R.

|

2008

|

tom Vol. 8, No. 4

196-202

EN

A method of sensitivity analysis for determining acceptable ranges of uncertainty of tested model's parameters was presented in this paper. Our approach is based on Morris method with modified sensitivity indicator. Solver from NuscaS system, which enables making simulation of casts solidification made from two-component alloys was used as a model. This model is built on the basis of the finite elements method, which is used to solve thermal conduction problem. Enthalpy formulation was used by the considered model. Acceptable ranges of parameters' uncertainty are determined in the way that parameters' values increases caused previously assumed changes of resulting quantity.

PL

W pracy przedstawiony jest sposób wykorzystania analizy wrażliwości do wyznaczenia dopuszczalnych przedziałów nie-pewności parametrów badanego modelu. Wykorzystywaną metodą jest metoda bazująca na metodzie Morrisa ze zmodyfikowanym wskaźnikiem wrażliwości. Badanym modelem jest solwer systemu NuscaS, który pozwala na przeprowadzanie symulacji krzepnięcia odlewów wykonanych ze stopów dwu-składnikowych. Model ten zbudowany jest w oparciu o metodę elementów skończonych, która wykorzystana jest do rozwiązania równania przewodzenia ciepła. Badany model wykorzystuje sformułowanie entalpowe do uwzględnienia zjawisk związanych z krzepnięciem. Dopuszczalne przedziały niepewności parametrów wyznaczane są w taki sposób, aby przyrosty wartości parametrów powodowały założone uprzednio zmiany wielkości wynikowej.

15

The use of Morris method in sensitivity analysis of a numerical model of solidification

70%

Sczygiol N. , Dyja R.

|

2007

|

tom Vol. 7, No. 2

250-254

EN

Numerical models which are created nowadays can be characterised by a high degree of complication. This complication concerns mainly big number of parameters of a model. As it turns out, not all parameters have equal influence on received results. Some of parameters have dominant influence, whereas influence of the other parameters is marginal and can be neglected with success. This means that part of parameters which are costly to determine and determination of which can require long-lasting research can be skipped in a given model without large influence on received results. And vice versa: some parameters which have dominant influence on results should be specified with the highest possible precision. Sensitivity analysis deals with the task of division of these parameters with regard on received results. In the sensitivity analysis we can distinguish between a local approach as well as a global approach. In case of the local approach the task is to investigate the influence of changes of individual parameter on received results with an assumption of unchangeability of other parameters. On the other hand, the global approach implies that the values of remaining parameters undergo changes, too. Many methods, varying among other things in computational time, belong to the global sensitivity analysis. The presented work concentrates on global sensitivity analysis, and particularly on one of method which belongs to global sensitivity analysis methods. Method used in this work is known in literature as Morris method. The most important feature of Morris method is its cheapness, which means the possibility to obtain the results with comparatively small effort of calculations. This feature enables its usage in case of models requiring qualification of many parameters, as well as in case when the models require many calculations. Such a situation takes place in case of numerical modelling of solidification process. And for this reason Morris method was decided to be used in sensitivity analysis of numerical model of solidification, because numerical modelling of solidification requires long time of calculations. In the presented work a solver that is part of Nuscas system is analysed. Nuscas is developed in the Institute of Computer and Information Sciences at Czestochowa University of Technology. The solver, which is used in numerical simulation of solidification process makes it possible to choose between explicit capacity and basic enthalpy formulation. It is also possible to choose the model of solid phase growth from models of equilibrium, nonequilibrium and intermediate growth of solid phase. The executed sensitivity analysis permits to rank given parameters with respect to their influence on received results. That knowledge allows distinction of particularly important parameters, as well as these parameters, which have little influence.

PL

W przedstawionej pracy zamieszczone jest omówienie analizy wrażliwości numerycznego modelu krzepnięcia za pomocą metody Morrisa. Analiza wrażliwości pozwala na badanie wpływu poszczególnych parametrów na uzyskiwane wyniki. Pozwala to ustalić parametry, które mają znaczny wpływ na wyniki, jak również te, których wpływ jest niewielki. Metoda Morrisa jest szczególnie przydatna do badania modeli, wymagających dużego nakładu obliczeń lub są opisane dużą liczbą parametrów, z powodu niewielkiej wymaganej przez nią wyk nanych symulacji komputerowych. Analiza wrażliwości numerycznego modelu krzepnięcia opracowanego przez autoró pracy jest przedstawiona jako przykład zastosowania metody Morrisa.

16

A Coherent Model of DCF Valuation

70%

Patena W.

Journal of Entrepreneurship, Management and Innovation

|

2011

|

tom 7

16-28

EN

Business valuation through DCF is recognized as one of the most popular valuation approaches. DCF valuation models, however, have become extremely complex. Modeling requires plenty of input data to be processed, the process is done in many stages, and the data obtained on each of the stages may be interrelated. The process then is not simply a chain of tasks. The modern models work via sophisticated mechanisms of loops being triggered whenever a new piece of information is revealed and the whole model needs updating. Technically speaking, in the spreadsheets environment, this may only be done with the use of iterations. The valuation model should also be subjected to the sensitivity analysis, which is able to quantify the impact of every single assumption made on the final company value. The analysis points out the set of critical assumptions, which have the major impact on the calculated company’s value. Apart from quantifying the impact of the assumptions, the analysis runs qualitative checks on the assumptions assessing the robustness of the arguments standing behind the critical factors for valuation. Consequently, the sensitivity analysis improves the objectivity of the model and mitigates the exposure for the possible results manipulation. The sensitivity analysis reveals its critical role in the valuation process and proves that it should be considered as the standard step in every DCF valuation.

17

The sensitivity analysis of even order biquadratic elliptic filters

70%

Pasko M. , Adrikowski T.

|

2006

|

tom Vol. 54, nr 1

101-105

EN

In this paper, the sensitivity analysis of the elliptic filters realized by using biquadratic structures was carried out. The influence f spread the structure parameter values on the shape of the frequency characteristic of the filter transmittance modulus was analyzed. The analysis was limited to the case of even order low-pass filter. Defining the proper class of the sensitivity coefficients, the changes influence !f individual structure parameters on the deviation of basic parameter values of the characteristic was considered. The considerations were ilustrated by the numerical example.

18

Estymowane modele równowagi ogólnej: zastosowanie metody dekompozycji funkcji do oceny zależności między parametrami postaci strukturalnej i zredukowanej

70%

Wróbel-Rotter R.

Zeszyty Naukowe Uniwersytetu Ekonomicznego w Krakowie

|

2013

|

nr 904

19-36

EN

The paper presents the application of high dimensional model representation to characterise the relationship between structural and reduced form coefficients of estimated general equilibrium models. The function representation is considered a state-dependent regression that is estimated non-parametrically, based on Monte Carlo sample, and generated from the probability distribution of structural parameters. The estimation method consists of recursive filtering and smoothing algorithms, derived from the Kalman filter, enhanced with special data re-ordering, to capture strong variability of the parameters in the state-dependent regression. The estimated function decomposition is used to build sensitivity indices. The methodology presented is illustrated with an example from the literature.

19

Using the analytic hierarchy process in evaluating decision alternatives

70%

Cabała P.

|

2010

|

tom Vol. 20, No. 1

5-23

EN

In this paper the method of the Analytic Hierarchy Process (AHP) is described. At the beginning the general assumptions of the method are characterized and discussed. These are related to assumptions held within General Systems Theory. Then the problems of pairwise comparisons of elements, with its use of a specific scale, as well as the resulting reciprocal matrix are presented. There are many ways of estimating the eigenvectors of this matrix. These eigenvectors reflect weights of preferences. Despite the fact that we are able to evaluate the consistency of judgements the problem of acceptable weights still remains. Therefore, by way of an illustration, the method for the sensitivity analysis of preferences is also discussed in the paper.

20

Sensitivity analysis for selective learning by feedforward neural networks

70%

Engelbrecht A. P.

|

2001

|

tom Vol. 45, nr 4

295-328

EN

Research on improving the performance of feedforward neural networks has concentrated mostly on the optimal setting of initial weights and learning parameters, sophisticated optimization techniques, architecture optimization, and adaptive activation functions. An alternative approach is presented in this paper where the neural network dynamically selects training patterns from a candidate training set during training, using the network's current attained knowledge about the target concept. Sensitivity analysis of the neural network output with respect to small input perturbations is used to quantify the informativeness of candidate patterns. Only the most informative patterns, which are those patterns closest to decision boundaries, are selected for training. Experimental results show a significant reduction in the training set size, without negatively influencing generalization performance and convergence characteristics. This approach to selective learning is then compared to an alternative where informativeness is measured as the magnitude in prediction error.