Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... GP populations tend to rapidly increase in size as the population evolves [13, 1, 33, 4, 26, 16, 32, 24] If unchecked, this consumes excessive machine resources and so is usually addressed either by enforcing a size or depth limit on the programs or by an explicit size component in the GP fitness [13, 12, 34, 29] although other techniques have been proposed [30, 6, 32, 31, 18] Depth or size limits [9, 20] and simple parsimony pressure [32] may have unexpected and untoward effects, while [11] shows that addition of duplicated code segments (i.e. addition of ineffective code, code that has no impact on the ....

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.

.... GP populations tend to rapidly increase in size as the population evolves [13, 1, 33, 4, 26, 16, 32, 24] If unchecked, this consumes excessive machine resources and so is usually addressed either by enforcing a size or depth limit on the programs or by an explicit size component in the GP fitness [13, 12, 34, 29] although other techniques have been proposed [30, 6, 32, 31, 18] Depth or size limits [9, 20] and simple parsimony pressure [32] may have unexpected and untoward effects, while [11] shows that addition of duplicated code segments (i.e. addition of ineffective code, code that has no impact on the ....

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.

.... 4, 30, 27, 25, 2, 41, 29] If unchecked this consumes excessive machine resources and so is usually addressed either by enforcing a size or depth limit on the programs or by an explicit size component in the GP tness measure which penalises larger programs, although other techniques may be used [13, 12, 42, 3, 36, 32, 39, 10]. Both main approaches have problems [13, 30, 38] 9, 20] Recently there has been increased interest in the underlying causes of bloat [28, 38, 26] It has been shown that the protective e ect of inviable code (which does not e ect the tness of the program) 27, 4] is not sucient to explain ....

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265-284. MIT Press, 1994.

....also be briefly mentioned. However it is almost impossible to make a fair comparison, because the results given in [22] are in diagram format only, and in any case, PIPE has ephemeral random constants while PEEL does not. The fitness cases were sampled at 101 equidistant points in the interval [0,10], 100,000 program evaluations were set for both PEEL and GGGP. The other settings for PEEL were population size 10, maximum depth 15, mutation coe#cient c mutation = 0.5, mutation intensity mi = 0.2, decay rate decay rate = 0.95, and the SSDT was refined every 50 generations. The settings for GGGP ....

....1921 1955 and significantly outperforms GP on the testing data 1956 1979. Actually the best individual found by PEEL generalizes much better on the data set 1956 1979 than any other method cited in [12] Much research has been done on parsimony pressure and control of individual complexity in GP [6, 19, 10, 25]. It is notable that, in our work, although no complicated parsimony mechanism is used, the results 12 still maintained impressively low complexity, furthermore obviously pre mature convergence was not observed. Although more accurate models of predicating sunspots have been obtained by GP [16] ....

H. Iba, H. de Garis, and T. Sato. Genetic programming using a minimum description length principle. In K. E. Kinnear, Jr., editor, Advances in Genetic Programming, pages 265--284. MIT Press, 1994.

....(iGP) 2] 6] 10] is a global, random search method for evolving genetic tree like programs. It is suitable for solving NPhard computational problems, many of which arise when we address inductive tasks. Previous works point out that iGP is successful for such tasks as system identification [3], symbolic regression [6] pattern recognition [16] scene analysis, robot navigation, concept learning [6] 8] time series prediction [3] etc. Inductive Genetic Programming is a specialization of the original Genetic Programming (GP) paradigm [6] for inductive learning [8] 12] The reasons ....

....problems, many of which arise when we address inductive tasks. Previous works point out that iGP is successful for such tasks as system identification [3] symbolic regression [6] pattern recognition [16] scene analysis, robot navigation, concept learning [6] 8] time series prediction [3], etc. Inductive Genetic Programming is a specialization of the original Genetic Programming (GP) paradigm [6] for inductive learning [8] 12] The reasons for this specialized term are: 1) inductive learning is essentially a search problem and GP is a versatile framework for exploration of ....

[Article contains additional citation context not shown here]

H. Iba, H. de Garis and T. Sato, 'Genetic Programming using a Minimum Descrip- tion Length Principle', In: Advances in Genetic Programming, K.Kinnear Jr.(ed.), The MIT Press, 265-284, 1994.

....similar size subtree of the other parent. By using these operators, code growth is considerably reduced without a ecting the performance of genetic programming. There are several studies that suggest taking into account the program size when computing the tness value. Iba, de Garis and Sato [6] de ne a tness function based on a Minimum Description Length (MDL) principle. The structure of the tree representing the genetic program is re ected in its tness value: mdl = Error Coding Length T ree Coding Length. Zhang and M uhlenbein [16] demonstrate the connection between accuracy and ....

....code growth in a robot guidance problem: 1) the straightforward editing out of irrelevant and redundant parts of code and (2) the use of a tness function that penalizes longer programs. They conclude that applying the penalty outperforms any kind of editing out, so providing new evidence for [6, 16]. Notwithstanding, other studies show that these seemingly irrelevant or redundant parts of code are useful because they shield the highly t building blocks of programs from the destructive e ects of crossover. Angeline [1] calls these apparently useless fragments of code introns, in analogy ....

Hitoshi Iba, Hugo de Garis, and Taisuke Sato, `Genetic programming using a minimum description length principle', in Advances in Genetic Programming, ed., Kenneth E. Kinnear, 265-284, MIT Press, (1994).

....the population size. 3.2 Method 1: using the tree size for selection Reducing the solution size is also an important topic in the study of GP. For example, we can put the description length as a penalty in the fitness function, and try to find solutions with the minimum description length (MDL) [Iba94]. In our study, however, we do not adopt this method, because a very small BDT may have a high fitness, but the correct classification rate may be very low. In our study, the first method (Method 1) for reducing the solution size is to select individuals according to their sizes. A size is ....

H. Iba, H. de Garis and T. Sato, "Genetic programming using a minimum description length principle," Advances in Genetic Programming, MIT Press (1994) 265283

....approaches such as neural networks) To reduce the solution size, some methods have been proposed in the context of GP study. For example, we can put the description length as a penalty term in the fitness function, and try to find solutions with the minimum description length (MDL) see [Iba et al.,1994]) This method cannot be used easily because we usually do not know how to choose the weight of the penalty term. In our study, we adopted the following method [Shirasaka et al., 1998] First, sort the individuals according to their fitness (recognition rate) and then sort the individuals with the ....

H. Iba, H. de Garis and T. Sato, "Genetic programming using a minimum description length principle," Advances in Genetic Programming, MIT Press (1994) 265-283

....but the size is smaller. Note that we may also adopt other (say, the roulette wheel) selection strategies. In that case, we must de ne the tness explicitly as a function of the decision rate and the tree size. For example, we may use the minimum description length (MDL) as the tness function [9]. Using this tness, however, we may nd a very small BDT, but the recognition rate may not be good. 4 Designing a Decision Tree for Character Recognition To show how to design a BDT using GP, and to make the discussion more concrete, we have applied it to a character recognition problem. The ....

.... cleaning, we can nd the global features as follows: Feature[ 1] Number of 4 Feature[ 2] Number of 3 Feature[ 3] Number of 2 Feature[ 4] Number of 1 Feature[ 5] Number of 0 Feature[ 6] Number of 1 Feature[ 7] Number of 2 Feature[ 8] Number of 3 Feature[ 9]=Number of 4 Feature[10] Percentage of the longest straight line Feature[11] Percentage of the 2nd longest straight line 4.2 The Evolution Parameters The evolution parameters are given as follows: The population size P= 200 The number of maximum generations = 500 The selection ....

H. Iba, H. de Garis and T. Sato, Genetic programming using a minimum description length principle, Advances in Genetic Programming, MIT Press (1994) 265-283

....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 exploits a set of the short term polynomials and a fitness function that penalize large network topology. However GMDH type training algorithms are performed well if noise and distortions of the training data are distributed by a Gauss low. In a presence ....

H. Iba, H. de Garis. Genetic Programming Using the Minimum Description Length Principle. In: K. Kinner (eds.), Advances in Genetic Programming, MIT Press, Cambridge, MA, 265-284, 1994.

....degree. In other words, the individual which is of lower degree and closer to the target time series has the higher possibility to be selected and inherited to the next generation. This fitness derivation is based on the MDL (Minimum Description Length) criterion, which was often used in GP (see [Iba94] and [Zhang95] for examples) When calculating the time series, some individuals may go overflow. In this case, the individual s fitness value gets so large that it will be weeded out from the population. We use several sets of time series as the training data for GP. This is to acquire the ....

Iba, H., deGaris, H., Sato, T., Genetic Programming using a Minimum Description Principle, in Advances in Genetic Programming, MIT Press, pp.265-284, 1994.

....degree. In other words, the individual which is of lower degree and closer to the target time series has the higher possibility to be selected and inherited to the next generation. This fitness derivation is based on the MDL (Minimum Description Length) criterion, which was often used in GP (see [Iba94] and [Zhang95] for example) When calculating the time series, some individuals may go overflow. In this case, the individual s fitness value gets so large that it will be weeded out from the population. We use several sets of time series as the training data for GP. This is to acquire the ....

Iba, H., deGaris, H., Sato, T., Genetic Programming using a Minimum Description Principle, in Advances in Genetic Programming, MIT Press, pp.265-284, 1994.

No context found.

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.

No context found.

H. Iba, H. de Garis, T. Sato, "Genetic Programming Using a Minimum Description Length Principle," Advances in Genetic Programming, vol. 1, pp. 265- 284, 1994.

No context found.

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265-284. MIT Press, 1994.

No context found.

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.

No context found.

Iba, H., de Garis, H., and Sato, T. (1994), "Genetic programming using a minimum description length principle," in Advances in Genetic Programming, K. E. Kinnear, Jr. (Ed.), Chapter 12, pp 265--284, MIT Press.

No context found.

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press, 1994.

No context found.

Iba, H., de Garis, H., and Sato, T. (1994). Genetic programming using a minimum description length principle. In Kinnear, Jr., K. E., editor, Advances in Genetic Programming, chapter 12, pages 265--284. MIT Press.

No context found.

H. Iba, H. de Garis, and T. Sato, "Genetic programming using a minimum description length principle," in Advances in Genetic Programming, Kenneth E. Kinnear, Jr. (Eds.), MIT Press, 1994, chap. 12, pp. 265--284.

No context found.

Hitoshi Iba, Hugo de Garis, and Taisuke Sato. Genetic programming using a minimum description length principle. In Kenneth E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265-284. MIT Press, 1994.

No context found.

Iba, H., de Garis, H., and Sato, T. (1994), "Genetic programming using a minimum description length principle," in Advances in Genetic Programming, K. E. Kinnear, Jr. (Ed.), Chapter 12, pp 265--284, MIT Press.

No context found.

H. Iba, H. de Garis, and T. Sato. Genetic programming using a minimum description length principle. In K. E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 12, pages 265-284. MIT Press, 1994.

No context found.

Iba, H., H. de Garis, and T. Sato: 1994, `Genetic Programming Using a Minimum Description Length Principle'. In: K. E. Kinnear, Jr. (ed.): Advances in Genetic Programming. MIT Press, Chapt. 12, pp. 265-284.

No context found.

H. Iba, H. de Garis, and T. Sato, "Genetic programming using a minimum description length Z. principle," in K. E. Kinnear, Jr. ed. , Advances in Genetic Programming, MIT Press: Cambridge, MA, 1994, Chapter 12, pp. 265#284.

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