C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 1(13):169--228, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the number of classi cation errors. In order to enable our system to minimize misclassi cation cost we modi ed the tness function, which is optimized by the evolutionary search process. Several EA based systems, which learn decision rules in either propositional (e.g. GABIL [5] GIL [10], EDRL [13] or rst order form (e.g. REGAL [9] were proposed. According to our knowledge, they are unable to process continuousvalued features directly and they cannot minimize misclassi cation costs. The remainder of the paper is organized as follows. The next section brie y discusses the ....

.... 250) 500) THEN Salary Amount Accept IF( 750) THEN Salary Purpose=carorhouse Accept (a) b) Fig. 3. Representation of rulesets: a) an example chromosome consisting of three strings, b) the corresponding ruleset. A similar approach was proposed by Janikow. His GIL system [10] conducts the search for rulesets using 14 operators. However, GIL is not able to handle continuous valued attributes directly, since it represents a condition as a sequence of binary ags corresponding to the values of an attribute (we use the same representation for nominal attributes) The ....

Janikow, C.: A knowledge intensive genetic algorithm for supervised learning. Machine Learning 13 (1993) 192-228.

.... (GAs) are randomized parallel search algorithms that search from a population of points [Holland, 1975, Goldberg, 1989] Current genetic algorithm based machine learning systems use rules to store past experience to improve their performance over time [Holland, 1975, Goldberg, 1989, Smith, 1985, Janikow, 1993, Grefenstette et al. 1990] However, many application areas, are more suited to a case based storage of past experience [Mostow et al. 1992, Huhns and Acosta, 1992, Sycara and Navinchandra, 1992, Goel and Chandresekaran, 1992] We propose and describe a system that uses a case base as a long ....

Janikow, C. Z. (1993). A knowledgeintensive genetic algorithm for supervised learning. Machine Learning, 13:189--228.

....in trying to solve every given problem. Genetic algorithms (GAs) are randomized parallel search algorithms that search from a population of points [4, 2] Current genetic algorithm based machine learning systems use rules to store past experience to improve their performance over time [4, 2] and [16, 6, 3]. However, many application areas, especially in the design domain, are more suited to a case based storage of past experience [13, 5] and [17, 1] We propose and describe a system that uses a case base as a long term knowledge store in a new genetic algorithm based design system that learns ....

Cezary Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....of syntactic constraints. 2.2.2 Selection and Genetic Operators We use the tournament selection method, with tournament size 2 and with a simple extension: if two individuals have the same fitness, the one with smaller complexity is selected. Complexity is measured by the following formula [12]: complexity = 2 x number of rules number of conditions. 1) This extension was motivated by observations in our experiments: sometimes the two individuals competing in the tournament had the same fitness value, even though they were different individuals. Once two individuals are selected ....

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning 13, 189-228. 1993.

....evolution. They have been applied to many optimization problems. The success of EAs is attributed to their ability to avoid local optima, which is their main advantage over greedy search methods. Several EA based systems, which learn decision rules in either propositional (e.g. GABIL [3] GIL [10], GA MINER [7] or first order (e.g. REGAL [8, 14] SIAO1 [1] form have been proposed. There are two key issues in our approach. The first one is the use of two nonstandard genetic operators, which we call changing condition operator and insertion operator. The second issue is the application of ....

Janikow, C.: A knowledge intensive genetic algorithm for supervised learning. Machine Learning 13 (1993) 192-228.

....techniques, which have been inspired by the process of biological evolution. The success of EAs is attributed to the ability to avoid local optima, which is their main advantage over greedy search methods. Several systems, which employ EAs for learning decision rules (e.g. GABIL [3] GIL [7], EDRL [8] were proposed. According to our knowledge all of them either work only with nominal attributes or discretize continuous valued ones prior to induction of rules using univariate methods. 2 The Weakness of the Univariate Discretization The univariate discretization methods, although ....

....condition, positive example insertion, negative example removal , rule drop are applied to a single ruleset RS c k (represented by chromosome) The other two: crossover and rule copy require two arguments RS ck 1 and RS ck 2 . A similar approach was proposed by Janikow. However, his GIL [7] system is not able to handle continuous valued attributes directly, since it represents a condition as a sequence of binary flags corresponding to the values of an at tribute (we use the same representation for nominal attributes) The changing condition is a mutation like operator, which ....

[Article contains additional citation context not shown here]

Janikow, C.Z.: A knowledge intensive genetic algorithm for supervised learning. Machine Learning 13 (1993) 192-228.

....the two search steps, namely the simultaneous search for threshold values for all continuous valued attributes and the discovery of decision rules. The search is performed by an evolutionary algorithm [6] EA) Several systems, which employ EAs for learning decision rules (e.g. GABIL [2] GIL [4], EDRL [5] were proposed. According to our knowledge all of them either use only nominal attributes or discretize continuousvalued ones prior to induction of rules. 2 Description of the method We assume that a learning set E = fe 1 ; e 2 ; e M g consists of M examples. Each example e 2 ....

Janikow, C.: A knowledge intensive genetic algorithm for supervised learning. Machine Learning 13 (1993) 192-228.

....individuals, which tends to make fitness computation more computationally expensive. In addition, it may require some modifications to standard genetic operators to cope with relatively complex individuals. Examples of GAs for classification which follow the Pittsburgh approach are GABIL [13] GIL [37], and HDPDCS [51] By contrast, in the Michigan approach the individuals are simpler and syntactically shorter. This tends to reduce the time taken to compute the fitness function and to simplify the design of genetic operators. However, this advantage comes with a cost. First of all, since the ....

....during the running of the algorithm. Hence, if we want to discover a set of classification rules predicting k different classes, we would need to run the evolutionary algorithm at least k times, so that in the i th run, i=1, k, the algorithm discovers only rules predicting the i th class [37], 43] The third possibility is to choose the predicted class most suitable for a rule, in a kind of deterministic way, as soon as the corresponding rule antecedent is formed. The chosen predicted class can be the class that has more representatives in the set of examples satisfying the rule ....

[Article contains additional citation context not shown here]

Janikow CZ. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning 13, 189-228. 1993.

....systems using this approach use the population as the model and have to introduce some methods for enforcing pluralistic solutions. Typically these systems use some kind of niching method in order to maintain several solutions in the population. The other approach, namely the Pittsburgh approach [8, 11, 3], uses a more complicated knowledge representation scheme and encodes the multimodality into the genetic individuals. Here the individuals are of varying length, typical is to have a set of fixed length bitstrings, and the whole solution is found in the individuals. Hence a GA that converges to a ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

.... orientations for the k nearest neighbor algorithm (Kelly Davis, 1991; Punch et al. 1993) to finite state automata (Fogel et al. 1966) and context free grammars (Wyard, 1991) to production system like rules (Booker et al. 1989; De Jong et al. 1993; Greene Smith, 1993, 1994; Holland, 1986; Janikow, 1993). Financial Forecasting Using Genetic Algorithms 549 Higher level constructs such as neural networks and LISP programs are very powerful representations. However, this power comes at the expense of an additional layer of complexity. Neural networks require search for an optimal topology as well ....

....applicability. Pittsburgh Versus Michigan Approach There have historically been two approaches to genetic classification, named after the universities at which the approaches originated: the Michigan approach (Booker et al. 1989; Holland, 1986) and the Pittsburgh approach (De Jong et al. 1993; Janikow, 1993; Smith, 1980, 1983) The main property distinguishing the two approaches is whether each population element represents a single classification rule or a set of rules. Although the two approaches have come with other accessories, we will use this single property to define and distinguish them. In ....

[Article contains additional citation context not shown here]

Janikow, C. Z. 1993. A knowledge-inten sive genetic algorithm for supervised learning. Machine Learning 13(2/3):189228.

....data and huge search spaces. Such limitations are avoided in Genetic Programming (GP) due to its stochastic search principle. The price to pay is that GP offers no direct way to incorporate expert knowledge, although the knowledge based issues of Evolutionary Computation are now recognized [Jan93]. In this paper, the emphasis is put on a particular, albeit rather general, expertise: in all application domains, variables most often have physical dimensions that cannot be ignored, for example, mass and length can not be added together. The restriction of the search space to dimensionally ....

....implies the useless generation of a vast majority of irrelevant trees [CY97] Several authors have addressed this problem using various kinds of bias. A first kind is provided by the expert through domain knowledge. The importance of taking this knowledge into account is now generally admitted [Jan93]. In an MD context, prior knowledge might concern the shape of the solution 2 . A significant improvement in the success rate of a GP application can be obtained by biasing the shape of the parse trees toward some shapes that are a priori judged interesting. This can be enforced by syntactic ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

.... and specialisation of rules [21] or adaptive operators, which work in a different way depending on the evolution level of the system ( 32] and [46] The Pittsburgh approach was initially proposed by Smith [72] Recent examples of it in a non fuzzy environment are the GABIL [21] and GIL [47] systems. This approach has not been applied to the generation of fuzzy rules for FRBCSs (see [10] 36] and [56] for some examples of its application in the design of Fuzzy Rule Based Control Systems) although its basic philosophy has been used in RB selection and tuning processes, as we will ....

Janikow, C.Z. (1993), "A knowledge intensive genetic algorithm for supervised learning," Machine Learning, Vol. 13, pp. 198-228.

.... of the Institute of Systems, Control, and Information Engineers (Japan) 205, 380, 996] Journal of the Operational Research Society, 352] Journal of the Society of Instrument and Control Engineers, 375, 509, 571] KI Lexikon, 190] Kikai Gijutsu Kenkyusho Shoho, 997] Machine Learning, [313, 386, 391, 516, 1065] Machine Learning Journal, 216] Mech. Syst. Signal Process. UK) 193] Methods of Information in Medicine, 881] Microprocessing and Microprogramming, 1073] Neural Computing and Applications, 847] Neural Network World, 802, 1004] New Electronics (UK) 799] New Scientist, 271, 661] ....

....[89, 90] Iwata, Tadashi, 747] Izui, Y. 497] Jacob, Christian, 148] Jacob, Varghese S. 471] Jacq, Jean Jose, 514, 515] Jaeger, E. P. 523, 525] Jain, Rajat, 790] Jakiela, Mark J. 173] Jakob, Wilfried, 1043] James, C. D. 367] Jang, Jyh Shing, 150] Janikow, Cezary Z. [516, 517] Janson, David J. 318] Janssen, M. 954] Javornik, B. 310] Jefferys, E. R. 518] Jenkins, W. M. 519] Jeon, Y. C. 1078] Jervis, M. 656] Jetzelsperger, R. 520] Jiang, Minga, 521] Jin, S. 885] Johnson, e.g. 239] Johnson, Glen E. 353] Joly, Georges, 56] Jones, A. ....

[Article contains additional citation context not shown here]

Cezary Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13(2-3):189--228, November-December 1993. y(CCA 855/94) ga:Janikow93a.

....GA runs for that class, showing that the GA converged to the same best rule despite variations in the random seed. 6 Related Work Several GAs designed for discovering some kind of comprehensible classification rules have been proposed in the literature. We briefly review some of them below. GIL [12] uses several generalization specialization operators proposed by [16] to extend the genetic operators of conventional GA, creating a knowledge intensive GA for the classification task. GIL follows the Pittsburgh s approach for rule learning, where each individual of the population corresponds to ....

Janikow, C.Z. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, v. 13, p. 189-228, 1993.

.... conditions) and numerical values (for example, multipliers in Bayesian approaches that measure the weight of the presence absence of a particular observation for or against a particular conclusion) Although, classifier systems have intensively been studied by genetic algorithm research [2] 21] [12], the rules learnt by classical classifier systems are purely symbolic and rely on two valued classical logic propositions are either true or false. Consequently, our research faces to some extend the challenge to learn a combination of symbolic and non symbolic parameters, which is not the ....

C. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," in Machine Learning, vol. 13 pp. 189-228, 1993.

....to represent the concepts in GA. In the Michigan approach [13, 25] each individual is represented by a fixed length string and corresponds to a partial concept description. The target concept is represented by the whole set of individuals in the population. Whereas in the Pittsburgh approach [4, 15, 22, 23], each individual is represented by a variable length string and corresponds to a whole target concept. The Michigan approach has the advantage that traditional genetic operators can be used without modification. However, various strategies have to be adopted in order to extract a non redundant ....

....involving highly irrelevant attributes will more likely be dropped. 2. 5 Adapting the Operator Probabilities While SIA selects the genetic operators with fixed probabilities, it is usually more helpful to dynamically adjust these probabilities based on the fitness of the individual chromosomes [15]. Take the mutation crossover operators as an example. While it is usually a good idea is to give the lessfit individuals a higher chance to mutate crossover, it may not be the case for those high fitness individuals. Hence, in ESIA, we adapt the probability pm for selecting the mutation ....

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13(23) :180--228, 1993.

.... The higher complexities of target problems usually mean that the representation of each competing individual is more complicated than just a string of 0 s and 1 s [27, 29] Some example representations of an individual in genetics based machine learning include an ifthen rule [27] a set of rules [96], a Lisp expression [29] or a vector of numbers [36] Because the structure of each individual can be complex, the reproduction operators such as mutation and crossover can also be more complex. In addition, more domain knowledge can be applied to create knowledge intensive reproduction [27, 94] ....

....or a vector of numbers [36] Because the structure of each individual can be complex, the reproduction operators such as mutation and crossover can also be more complex. In addition, more domain knowledge can be applied to create knowledge intensive reproduction [27, 94] such as those used in GIL [96]. Since the goal is to develop a heuristic or strategy for problem solving, this area of evolutionary computing also has to deal with noisy conditions more often. This condition means that the fitness of each individual may not be exact, and that multiple applications of each individual may ....

[Article contains additional citation context not shown here]

C. Z. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," Mach. Learn., vol. 13, pp. 189--228, November/December 1993.

.... (GAs) are randomized parallel search algorithms that search from a population of points (Holland 1975; Goldberg 1989) Current genetic algorithm based machine learning systems use rules to store past experience to improve their performance over time (Holland 1975; Goldberg 1989; Smith 1985; Janikow 1993; Grefenstette, Ramsey, Shultz 1990) However, many application areas, especially in the design domain, are more suited to a case based storage of past experience (Mostow, Barley, Weinrich 1992; Huhns Acosta 1992; Sycara Navinchandra 1992; Goel Chandresekaran 1992) We propose and ....

Janikow, C. Z. 1993. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning 13:189--228.

....for correspondence: LMS and CMAP, Ecole Polytechnique, 91128 Palaiseau Cedex, France M. Sebag, M. Schoenauer and M. Peyral Revisiting the Memory of Evolution 2 optimization [36] or shape design [44] see [35] for general recommendations about representation operators) As noted by Janikow [26], this transition is quite similar to what happened in the field of artificial intelligence (see [41] for a survey) at first, people were fascinated by the generality of the principles at hand and they aimed at universal tools, e.g. the General Problem Solver [33] Afterward, they realized that ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....assume that significant adjustment of the fitness function, representation and other parameters will be required to solve any interesting and thereby non trivial problem. Ces remarques sont tr es voisines de celles qui concernaient il y a quelques d ecennies les r esolveurs de probl emes en IA [32] : une phase d emerveillement devant les capacit es d un outil universel est suivie d une seconde phase, plus avertie, o u l on red ecouvre l int eret de tenir compte des particularit es des probl emes trait es. Le plus grand d efaut des AG est sans conteste leur lenteur. Une cause identifi ee de ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....In most of the previous efforts to apply evolutionary computation to concept learning, binary string representations have been evolved with a genetic algorithm and mapped into some form of symbolic representation for evaluation, such as propositional logic. For some examples of this approach, see [4, 1, 3]. In the work described here, we take a different approach by experimenting with a biologically inspired representation in which concept descriptions are evolved using a model of the immune system. For other approaches to evolving models of the immune system, see the pioneering work of Forrest et ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13(2/3):189--228, 1993.

....show that the methodology can lead to better results, as well as to clear savings in computational effort, compared to learning with purely inductive GAs. Keywords: Background Knowledge, Genetic Algorithms, Inductive Learning. 1 Introduction Systems based on genetic algorithms (GAs) like [6, 2, 5, 10], have been shown to be an alternative to more traditional concept learners in cases where there is little or no domain theory available. There are also some studies, e.g. 7, 10] showing how the search in GAs can be constrained and guided through additional knowledge about the search space, but ....

....Genetic Algorithms, Inductive Learning. 1 Introduction Systems based on genetic algorithms (GAs) like [6, 2, 5, 10] have been shown to be an alternative to more traditional concept learners in cases where there is little or no domain theory available. There are also some studies, e.g. [7, 10], showing how the search in GAs can be constrained and guided through additional knowledge about the search space, but overall there has been little work done in integrating background knowledge into concept learning GAs. However, in cases where there exists some background knowledge, pure ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....in the second class, we want to see just the opposite, i.e. the reflectance at the first wavelength should be lower than the reflectance at the second wavelength. The above approach or similar approaches combining multiple learning criteria can be found in the literature (De Jong et al. 1993; Janikow 1993). We have applied this approach, in particular, to the problems of breast cancer prognosis (see Section 8.4.3) A similar winner take all approach was used in classification of the iris data (see Section 8.4.1) however the merit function value was calculated using equation (7.2) instead of ....

Janikow, C.Z. (1993). "A Knowledge-Intensive Genetic Algorithm for Supervised Learning." In Genetic Algorithms for Machine Learning, pp.33-72. Edited by Grefenstette, J.J. Massachusetts: Kluwer Academic Publishers.

....new rules. In some cases, variable length classifier sets are used, employing modified genetic operators for dealing with these variable length and position independent genomes. This model was initially proposed by Smith in 1980 [38] Recent instances of this approach are the GABIL [13] and GIL [31] systems. As mentioned in [12] the Michigan approach will prove to be most useful in an online, real time environment in which radical changes in behaviour cannot be tolerated, whereas the Pittsburgh approach will be more useful for off line environments in which more leisurely exploration and ....

Janikow, C.Z., A Knowledge Intensive Genetic Algorithm for Supervised Learning. Machine Learning 13 (1993) 198-228.

....paradigm) can also turn successful evolutionary optimization into disasters: see the problem of identification of constitutive laws of hyper elastic materials (section 3.4. 3) or all applications where standard GA failed until a Behavioral Memory based process was used (section 6) Following [85], we would compare the present situation of EC to that of AI regarding problem solvers some years ago: during a first enthusiastic phase, people had been looking for the general problem solver that would address all possible problems; it progressively occurred that this was merely a mirage, and ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....Michigan type systems, see for example [9, 14] use a fixed length chromosome representation. Most often a chromosome represents a single rule and hence these systems have to develop special methods in order to be able to deal with disjunctive concepts. In the Pittsburgh approach, see for example [5, 10, 15], the situation is reversed. These systems use chromosomes of varying length, where each chromosome encodes a whole rule based classification theory. Therefore these systems can adapt to disjunctive concepts naturally, at the cost of having more complex chromosomes and genetic operators. JGA ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....PROpositional Logic) is an inductive and supervised learning system of logic rules using a genetic algorithm for searching the space of theories explaining the set of examples. The system is similar in spirit to BEAGLE [7] and GABIL [5] and other genetic algorithms for inducing logic rules [1, 3, 6, 10]. However, the rules generated by EVOPROL are always logic rules with logical connectives such as AND, OR and NOT. This approach results in more intelligible rules because, as opposed to BEAGLE, EVOPROL does not mix data types. As opposed to GABIL and other systems [3] we can handle both ....

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervized learning. Machine Learning, 13:189--228, 1993.

....functions repeatedly during search for the best complex that also passes some minimum treshold of reliability until no more reliable complexes can be found. GIL Janikow s Genetic Based Inductive Learning system (GIL) is a knowledge intensive genetic algorithm approach for supervised learning (Janikow, 1991; 1993). Genetic algorithms (GAs) are adaptive methods of searching a solution space by applying operators modeled after the natural genetic inheritance and simulating the Darwinian struggle for survival of the fittest. In general, a genetic algorithm (GA) performs a multi directional search and ....

Janikow, C.Z. (1993). A Knowledge-Intensive Genetic Algorithm For Supervised Learning.

....mutation provides new rules. In some cases, variable length classifier sets are used, employing modified genetic operators for dealing with these variable length and position independent genomes. This model was initially proposed in [31] Recent instances of this approach are the GABIL [7] and GIL [25]. An interesting discussion about the problems generated by the use of these approaches can be seen in [19] In the recent literature we may find different algorithms that use a new learning model based on GAs called the iterative rule learning approach. In this approach, the GA provides a partial ....

Janikow C.Z., A Knowledge-intensive Genetic Algorithm for Supervised Learning. Machine Learning, 13, pp. 189-228. (1993).

.... i.e. a tuple is assigned the class of the nearest prototype, according to a given distance metric. GIL uses several generalization specialization operators proposed by [Michalski 83] to extend the genetic operators of conventional GA, creating a knowledge intensive GA for classification tasks [Janikow 93] REGAL learns first order logic (FOL) class descriptions [Neri Giordana 95] However, it assumes that the user provides a kind of template of the logical formula to be learned. This reduces the autonomy of the system, which is a serious drawback in the context of DM. SIAO1 also learns FOL ....

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13, 1993, 189-228.

....a inductive and supervised learning system of first order predicate logic rules using a genetic algorithm for searching the space of theories explaining the set of examples. The system is similar in spirit to BEAGLE [17, 18] and GABIL [12, 14] and other genetic algorithms for inducing logic rules [1, 4, 15, 24]. However, the rules generated by EFOPREL are always first order predicate logic rules with logical connectives such as AND, OR and NOT. EFOPREL results in more intelligible rules because, as opposed to BEAGLE, EFOPREL does not mix data types. As opposed to GABIL and other systems [4] EFOPREL can ....

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervized learning. Machine Learning, 13:189--228, 1993.

....systems that may be adapted to different situations. Genetic Algorithms (GAs) are known to be adaptive search mechanisms that can investigate large search spaces and point out interesting structures in these spaces. GAs have been used successfully in learning both propositional concepts, e.g. [17, 5, 10, 11, 32], and relational concepts, e.g. 8] In this article we investigate the adaptability of GAs to rule based concept learning in different kinds of propositional learning situations and compare them to other propositional concept learners. We utilize a two level approach in learning disjunctive ....

....approach the whole rule set is encoded as an individual. As a consequence the individuals have varying lengths and a complete solution can be found in one single individual. This coding scheme has been used by various researchers in various systems, e.g. in GABIL [5] SAMUEL [11] and GIL [17]. With its fix length individuals the Michigan approach can use the standard genetic operators, and thus draw experience from the years of theoretical and comparative research put into these. Individuals in Pittsburgh type systems are more complex, which leads to more complex and non standard ....

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

....paper we presented GIGA, a novel approach to classification that employs a genetic search algorithm in order to approximate ideal concept descriptions to be used by a nearestneighbor classifier. Our algorithm follows the Pittsburgh approach to machine learning oriented GAs [Smi83, Koz91, JSG93, Jan93] in that each individual of the population encodes a complete solution of the classification problem. But in contrast to other systems that learn explicit abstractions from examples (e.g. decision trees) our instance based concept description permits a scalable output representation language ....

C.Z. Janikow. A knowledge intensive genetic algorithm for supervised learning. Machine Learning, 13:198--228, 1993.

.... (GAs) are randomized parallel search algorithms that search from a population of points (Holland, 1975; Goldberg, 1989) Current genetic algorithm based machine learning systems use rules to store past experience to improve their performance over time (Holland, 1975; Goldberg, 1989; Smith, 1985; Janikow, 1993; Grefenstette et al. 1990) However, many application areas, especially in the design domain, are more suited to a case based storage of past experience (Mostow et al. 1992; Huhns and Acosta, 1992; Sycara and Navinchandra, 1992; Goel and Chandresekaran, 1992) Typically, a genetic algorithm ....

Janikow, C. Z. (1993). A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228.

No context found.

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 1(13):169--228, 1993.

No context found.

C. Z. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," Mach. Learn., vol. 1, no. 13, pp. 169--228, 1993.

No context found.

Janikow, C. Z. #1993#. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning 13, 189#228.

No context found.

C. Z. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," Mach. Learn., vol. 1, no. 13, pp. 169--228, 1993.

No context found.

Cezary Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

No context found.

Cezary Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

No context found.

Cezary Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189--228, 1993.

No context found.

C. Z. Janikow, `A knowledge-intensive genetic algorithm for supervised learning', Machine Learning, 1(13), 169--228, (1993).

No context found.

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning, 13:189-228, 1993.

No context found.

C. Z. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," Mach. Learn., vol. 13, pp. 189--228, 1993.

No context found.

C. Z. Janikow, "A knowledge-intensive genetic algorithm for supervised learning," Mach. Learn., vol. 1, no. 13, pp. 169--228, 1993.

No context found.

C.Z. Janikow. A knowledge intensive genetic algorithm for supervised learning. Machine Learning, 13:198-228, 1993.

No context found.

C.Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Machine Learning 13, 189-228. 1993.

No context found.

C.Z. Janikow, "A Knowledge-Intensive Genetic Algorithm for Supervised Learning," Machine Learning, vol. 13, pp. 189-228, 1993.

No context found.

C. Z. Janikow. A knowledge-intensive genetic algorithm for supervised learning. Mach. Learn., 13(2/3):189--228, November/December 1993.

No context found.

C. Z. Janikow, A knowledge-intensive genetic algorithm for supervised learning, Machine Learning 13, no. 2-3 (1993) 189--228.

First 50 documents  Next 50

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