H. R. Madala and A. G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, LLC, January 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....[9] is a simple linear regression based machine learning algorithm designed to learn a mapping from discrete valued inputs to continuous valued outputs. It operates by searching for useful conjuctive queries that form indicator functions for a linear regression. It is very similar to GMDH modeling [6], Model Trees [11] or even stepwise polynomial regression. In a dataset with two input attributes Gender and HairColor and one output attribute Age , a typical RADREG model might decide to use the following features: Gender=Female AND HairColor=Grey) HairColor=Red) Gender=Male AND ....

H. R. Madala and A. G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, LLC, January 1994.

....neural networks. The developed methods were successfully used to optimize the set of initial variables. The use of variables, the developed methods selected, results in an improvement of the prediction ability of the neural network. To avoid over fitting, a Group Method of Data Handling (GMDH) [7] has been suggested which allows to generate neural networks of appropriate complexity. To learn networks to generalize well, the GMDH exploits a regularity criterion calculated on the training and validating examples. The GMDH type neural networks were successfully applied to real world problems ....

Madala, H. R. and Ivakhnenko, A. G.: Inductive Learning Algorithms for Complex Systems Modeling, CRC Press Inc., Boca Raton, 1994.

....Typically the data is split into two groups. One data group is used to train the network and the other data group is used to rank the nodes to determine which nodes survive to form the input to the next layer. The GMDH can thus be seen as a methodology for distributed self organizing computation [11] [12] 13] 4 CARTESIAN GENETIC PROGRAMMING Cartesian Genetic Programming [14] 15] 16] 17] 18] as illustrated in figure 2) is a new form of Genetic Programming in which a program is represented as an indexed directed graph. The directed graph is encoded in the form of a linear string of ....

H. R. Madala and A. G. Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, 1994.

....Group Method of Data Handling (GMDH) was first proposed by Alexy G. Ivakhnenko [4] The GMDH network topology has been traditionally determined using a layer by layer pruning process based on a pre selected criterion of what constitutes the best nodes at each level. The traditional GMDH method [5] [6] is based on an underlying assumption that the data can be modelled by using an approximation of the Volterra Series or Kolmorgorov Gabor polynomial as shown in equation(1) y = a 0 a i x i a ij x i x j (1) k=1 a ijk x i x j x k . 2.1 GMDH Layers When constructing a GMDH ....

H. R. Madala and A. G. Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, 1994.

....Group Method of Data Handling (GMDH) was first proposed by Alexy G. Ivakhnenko [3] The GMDH network topology has been traditionally determined using a layer by layer pruning process based on a pre selected criterion of what constitutes the best nodes at each level. The traditional GMDH method [4] [5] is based on an underlying assumption that the data can be modelled by using an approximation of the Volterra Series or Kolmorgorov Gabor polynomial as shown in equation (1) y = a 0 a i x i a ij x i x j (1) k=1 a ijk x i x j x k . 2.1 GMDH Layers When constructing a GMDH ....

H. R. Madala and A. G. Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, 1994.

....Group Method of Data Handling (GMDH) was first proposed by Alexy G. Ivakhnenko [4] The GMDH network topology has been traditionally determined using a layer by layer pruning process based on a pre selected criterion of what constitutes the best nodes at each level. The traditional GMDH method [5] [6] is based on an underlying assumption that the data can be modeled by using an approximation of the Volterra Series or Kolmorgorov Gabor polynomial as shown in equation(1) y = a 0 a i x i a ij x i x j (1) k=1 a ijk x i x j x k . 2.1 GMDH Layers When constructing a GMDH ....

H. R. Madala and A. G. Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, 1994.

....process subtask as well. In this way, data transformation, dimension reduction, variables selection (preprocessing) data mining, and validation of models [4, 5] flow together into a single, autonomously running process. Methods SODM is built on the Group Method of Data Handling (GMDH) approach [1, 6] and can be composed of different self organizing modeling technologies. We mainly used GMDH here, too, to generate parametric regression models, but also Fuzzy Rule Induction (FRI) for creation of rules of linguistic terms and Analog Complexing based classification (AC class) for providing a ....

Madala, H.R., Ivakhnenko, A.G.: Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, Boca Raton 1994

....A.F. Hsaxaeato F.A. http: www.gmdh.net AHHOTAImI PacCMaTpHBaeTc YIpHMeHeHHe 9KClIeplIMeHTaJIbHblX jIaHHblX. 1. PIIII III 1TpKTIIqIrIX 3q H ppOTK TOpTK BOOOB MTO pyOBOFO qT Apeos (MFYA) asena nossnea mooro cnea ssmacnenssx ropuos, abifi a3 KOTOpblX eaaaqea oeeneaabix ycnoaafi aueaea [1,2,3,4,5,6]. Bbi6op ropMa 3aBHC KaK OT TOqHOGTH H OHOTBI HHOpMaRHH, ecTaeHHO B Bb6ope 9KcepHMeHTbHblX aHHblX, TaK HOT BHa pemaeMofi 3aaqa. aHHbI o63op HMeeT uebm aTb ropus MFYA paqasx caes aueaea. peBapebHO onameM OCHOBHble qepTbI paccMaasaeMbX ropMOB. 1.1. Houuou,u onopu yugqu Mevo ocaosaa aa nepe6ope, ....

....v.e. noceosavesaou OO6OBaH Moeefi, Bb6aeMbX 3 MHoeCTBa Moeefi KaaTOB no 3aaHHoMy Tepm. OqTU BCe U3BeCTHbe ropMb YA UCnOb3ymT OUHOMUbHble onopube yHKUU. O6 CBfl3B Mey BXOHBIM BBIXOHBIM epeMeHHblMU HaxouTCfl B Be yHKUOHbHoro pa Bosveppa, acevasfi aaor ovoporo assecvea a noaaou Kouoroposa Fa6opa [6]: M M M M M M y :a0 Za, x, ZZao. x,x ZZZao.x,xx, i= i= j= i= j= re X(x ,x 2 ) sevop sxoasx nepeueaasx; A (a,a2, a) sevop oeauaeavos caraeusx. HpMemTCa pe onopHbe yHK, HaMep rapMoHuece orcTuece: M a Y = a 1 exp( x i ) COHOCTB cybI Moeu oeuuBaeTcfl nO qucy UCnOb3yeMbX ....

[Article contains additional citation context not shown here]

Madala,H.R. and Ivakhnenko,A.G., Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton, 1994.

.... optimality and use heuristic search to nd a good, if not optimal, rule This reasonable heuristic has been used to good e ect in several rule and decision list induction algorithms such as CN2 [ Clark and Niblett, 1989 ] and PRIM [ Friedman, 1998 ] and stepwise regression analysis such as [ Madala and Ivakhnenko, 1994 ] Hill climbing is simple: Let ru 1 = best rule of size 1 (call it att 1 = val 1 ) Let ru 2 = best rule of the form ru 1 att 2 = val 2 . Let ru k = best rule of the form ru k 1 att k = val k . Then use ru k as the approximate argmax of Equation 1. In subsequent experiments we will ....

....are almost always substantially better than those learned by Hill climbing, especially at the lowcoverage end of the curve. 5. 1 RADREG RADSEARCH gives us an excellent opportunity to nd good additive models using the same kind of stepwise linear regression as MARS [ Friedman, 1988 ] GMDH [ Madala and Ivakhnenko, 1994 ] or Projection if edunum 10 marital=NeverMarried relation =child then predict wealth=poor (99.5 testset agreement) else if marital=MarriedCivil job=Professional then predict wealth=rich (70.8 testset agreement) else if. Table 5: A fragment of a decision list. if ....

H. R. Madala and A. G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, LLC, January 1994.

....that fully connected neural networks learnt from data can not be represented in a readable form due to a large number of connections between the input, hidden and output neurons. Group Method of Data Handling (GMDH) based on a polynomial theory of complex systems has been invented by Ivakhnenko [6, 7, 8]. The GMDH does not require to preset the neural network structure and allows to comprehensively present a classification rule as a concise set of short term polynomials. To improve generalization ability of the GMDH type networks, the authors [9, 10, 11] used a genetic inductive approach, which ....

....technique that requires to divide the dataset into the training, testing and validating subsets. II.B. GMDH Type Neural Networks In contrast to the FNNs, the GMDH type neural networks do not need to preset their structures and may be comprehensively described by a concise set of polynomials [6, 7, 8]. The GMDH type networks are the multi layered ones consisted of Fig. 1: An example of polynomial network The neuron candidates that were selected at each layers are depicted as grew boxes. A neuron y2 that provides the best classification accuracy assigns to be an output neuron. A resulting ....

[Article contains additional citation context not shown here]

H.R. Madala, A.G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton, 1994.

....that fully connected neural networks learnt from data can not be represented in a readable form due to a large number of connections between the input, hidden and output neurons. Group Method of Data Handling (GMDH) based on a polynomial theory of complex systems has been invented by Ivakhnenko [6, 7, 8]. The GMDH does not require to preset the neural network structure and allows to comprehensively present a classification rule as a concise set of short term polynomials. To improve generalization ability of the GMDH type networks, the authors [9, 10, 11] used a genetic inductive approach, which ....

....technique that requires to divide the dataset into the training, testing and validating subsets. II.B. GMDH Type Neural Networks In contrast to the FNNs, the GMDH type neural networks do not need to preset their structures and may be comprehensively described by a concise set of polynomials [6, 7, 8]. The GMDH type networks are the multi layered ones consisted of the neurons whose transfer function g is a short term polynomial. For example, a linear polynomial is y = g(u 1 , u 2 ) w 0 w 1 u 1 w 2 u 2 , 1) where u 1 and u 2 are the variables, and w 0 , w 1 , w 2 are the neuron weights ....

[Article contains additional citation context not shown here]

H.R. Madala, A.G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton, 1994.

....is transformed to a conditional form given by Gauss, the various interpolation problems of artificial intelligence, such as pattern recognition, dependence detection, stepwise prediction of random processes, etc. can be solved by general algorithms, mostly by the combinatorial GMDH algorithm [1, 2]. These algorithms differ mostly in the choice of output variables and modeling space coordinates. For pattern recognition and dependence detection, the discriminant functions are found, whereas for prediction, the dependence of the further values of one of the variables on the current and delayed ....

....MAPE = 31.75 . We can conclude that filtration proves to yield more accurate results here and, moreover, the obtained model includes no primary features, but only their pair covariance. 4. USING THE GMDH ALGORITHM FOR FILTERING A DATA SAMPLE A similarity between the combinatorial GMDH algorithm [1, 2] and the Kalman filter [9] consists in the fact that they both use a simplified model of an object. The difference is that, in the Kalman filter, a model is given by a human, an author of filtration, whereas in the GMDH algorithm, it is obtained by a self organization, i.e. by sorting out on the ....

Madala, H.R. and Ivakhnenko, A.G., Inductive Learning Algorithms for Complex Systems Modeling, Boca Raton: CRC, 1994.

....problems. This is the base for the general approach for solving different problems. Algebraic NEURAL NETWORKS transformations can be supplemented by estimating the efficiency of primary and secondary arguments candidates [7, 9] and by minimizing their number by the combinatorial GMDH algorithm [10] in all interpolation problems. In addition, the GMDH algorithms are applied in order to increase the generalization property of the decision function, which is necessary in the case of incomplete input data samples. 1. THE CONCEPT OF PHYSICAL CLUSTERIZATION According to this concept, for any ....

Madala, H.R. and Ivakhnenko, A.G., Inductive Learning Algorithms for Complex Systems Modeling, Boca Raton:

....is developed, which has polynomial complexity. It means that computation time is proportional to number of input variables. Polynomial complexity is necessary to meet challenge of nowadays increase of number of input variables in data samples. To reach polynomial complexity of GMDH algorithms [3] input variables should be preliminary evaluated on their efficiency on information level. Inductive sorting of models candidates begins from the most efficient input variables. Complexity of models is steadily increased until minimum of external error criterion will be reached. Three rules of ....

Madala H.R., Ivakhnenko A.G. Inductive Learning Algorithms for Complex Systems Modeling, CRC Press Inc., Boca Raton, 1994, p.384.

....et al. generating the set of models candidates and for estimating them according to the external criterion. Until now, a very limited number of different reference functions are employed. Most researchers use a linear, with respect to the coefficients, polynomial as the reference functions [9]. They justify it by the fact that this polynomial is a discrete analog of the general function decomposition into the Volterra series, although only a small part in the beginning of this series is actually used. Several examples prove that, for most modeling tasks, the more efficient models are ....

.... that, for most modeling tasks, the more efficient models are obtained for the fractionally polynomial reference functions considered in [10] A separate line of modeling uses harmonic and exponential reference functions, but only the cases with the small number of time variables are considered [9]. The problem of choosing reference functions is still to be considered. Another insufficiently developed problem consists in separating data into learning and test parts in order to obtain the external criterion of the model evaluation. It is stated in [12] that an optimal (according to any ....

Madala, H.R. and Ivakhnenko, A.G., Inductive Learning Algorithms for Complex Systems Modeling, Boca Raton: CRC, 1994.

....included in the observations and there are enough observations in data sample; it is possible that a part of past behavior will be repeated. This approach is recommended when number of input variables is bigger than number of observations, in opposite case parametric GMDH algorithms may be used [3]. If we succeed in finding for the last part of behavior trajectory (starting pattern) one or more analogous parts in the past (analogous pattern) the prediction can be achieved by applying the known continuation of these analogous patterns (Fig. 1) The point A nearest to each given output ....

Madala,H.R. and Ivakhnenko,A.G. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton, 1994.

....and narrowing of a complete set of variables in the layers of a neural neO. 1. Introduction Solving practical problems and developing theoretical questions of the Group Method of Data Handling (GMDH) produced a broad spectrum of computing algorithms, each designed for a specific application [1 6]. The choice of an algorithm depends both on the accuracy and completeness of information presented in the sample of experimental data and on the type of the problem to be solved. The purpose of this review is to demonstrate GMDH algorithms for different applications. We will first describe basic ....

....models according to a specified criterion. Nearly all known GMDH algorithms use polynomial support functions. General connection between input and output variables can be found in the form of a functional Volterra series, whose discrete analogue is known as the Kolmogorov Gabor polynomial [6], M M M M M M i=l i=l j=l i=1 j=l k=l where P( x,x2, xM) is the vector of input variables and .4(a,a2, aM) is the vector of the summand coefficients. Other support functions are also used, e.g. harmonic or logistic ones: Y =a0 1 exp( x i) The complexity of the model structure is ....

[Article contains additional citation context not shown here]

Madala, H.R. and Ivakhnenko, A.G., Inductive Learning Algorithms for Complex Systems Modeling, Boca Raton: CRC Inc., 1994.

No context found.

H. R. Madala and A. G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press, LLC, January 1994.

No context found.

H. R. Madala, A.G. Ivakhnenko (1994). Inductive Learning Algorithms for Complex Systems Modelling, CRC Press Inc., Boca Raton.

No context found.

H.R. Madala, A.G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton, 1994.

No context found.

Madala H.R., Ivakhnenko A.G. 1994. Inductive Learning Algorithms for Complex System Modeling. CRC Press Inc., Boca Raton.

No context found.

Madala, H.R. and Ivakhnenko, A.G., Inductive LearningAlgorithm for Complex Systems Modeling, CRC Press, 1994.

No context found.

Madala, H.R. and Ivakhnenko A.G. (1994) Inductive Learning Algorithms for Complex Systems Modeling. CRC Press Inc., Boca Raton.

No context found.

H.R.Madala and A.G.Ivakhnenko, Inductive Learning Algorithms for Complex Systems Modeling, Boca Raton, CRC Press, Inc., 1994.

No context found.

H.R. Madala, A.G. Ivakhnenko. Inductive Learning Algorithms for Complex Systems Modeling, CRC Press Inc., Boca Raton (1994).

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC