D. P. Greene and S. F. Smith, "Competition-based induction of decision models from examples," Machine Learning, vol. 13, pp. 229--257, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....search space. The selection scheme operationalizes exploitation and recombination effects exploration. Goldberg provides a thorough account of the mechanics of genetic search. Considering the present problem context, each population member can specify a model expressed in symbolic rule form [7, 12, 26] or a weight vector on the predictor variables. 19] This article takes the latter approach and considers models that express a linear combination of attributes (predictors) Such linear models are often preferred for decision making, given the ease of interpretation of results and higher ....

....more sophisticated representations. Rule based representations capable of discerning complex patterns in the data [3, 15, 23] can be particularly advantageous. Genetic algorithms have been successfully applied in learning rules comprising of logical combinations of attribute value restrictions. [7, 12 14] Such nonlinear representations can be expected to exhibit superior performance across file depths, and models expressed in rule form also possess the advantage of direct interpretability. Other nonlinear forms, like the parse tree representations used in genetic programming, 20] are also ....

D.P. GREENE and S.F. SMITH, 1993. Competition Based Induction of Decision Models from Examples, Machine Learning 13, 220 -- 257.

....it to the [0. 1] interval for the basic fitness function. The calculation of fitness was modified in two ways: 1. The evaluation of the population was made into a twostep process to encourage the GA to find multiple rules. This modification is similar to the approach described by Greene Smith [6]; strings only get to count themselves as explaining positive examples that have not already been explained by a higher fitness string. The process is illustrated in Figure 2. 2. A parameter was added to specify the minimum number of examples that should be described by each string; strings below ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13(2/3):229--257, November/December 1993.

....We can avoid the need for niching in the Michigan approach by running the GA several times, each time discovering a different rule. The drawback of this approach is that it tends to be computationally expensive. Examples of GAs for classification which follow the Michigan approach are COGIN [30] and REGAL [28] So far we have seen that an individual of a GA can represent a single rule or several rules, but we have not said yet how the rule(s) is(are) encoded in the genome of the individual. We now turn to this issue. To follow our discussion, assume that a rule has the form IF cond 1 ....

....operators [43] 3.1.3 Representing the Rule Consequent (Predicted Class) Broadly speaking, there are at least three ways of representing the predicted class (the THEN part of the rule) in an evolutionary algorithm. The first possibility is to encode it in the genome of an individual [13] [30] possibly making it subject to evolution. The second possibility is to associate all individuals of the population with the same predicted class, which is never modified during the running of the algorithm. Hence, if we want to discover a set of classification rules predicting k different ....

Greene DP & Smith SF. Competition-based induction of decision models from examples. Machine Learning 13, 229-257. 1993.

....have the capability of learning several rules, since it is often difficult to tell in advance if one or more rules are needed in the classification. In concept learning Genetic Algorithms (GAs) the problem of multimodality has been attacked by two different approaches. The Michigan approach, e.g. [13, 5, 7], uses fixed length bitstrings as individuals, where each individual represents one rule. In multimodal cases the 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 ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

....it to the [0. 1] interval for the basic fitness function. The calculation of fitness was modified in two ways: 1. The evaluation of the population was made into a twostep process to encourage the GA to find multiple rules. This modification is similar to the approach described by Greene Smith [6]; strings only get to count themselves as explaining positive examples that have not already been explained by a higher fitness string. The process is illustrated in Figure 2. 2. A parameter was added to specify the minimum number of examples that should be described by each string; strings below ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13(2/3):229--257, November/December 1993.

.... function (Rendell, 1990) to weights and 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 ....

....Sikora and Shaw (1994) in their genetic classification system. Sequential niching repeatedly runs a traditional GA, each time making sure that the population searches a new area of the space. Two previous learning systems that utilize implicit niching methods also deserve mention. In both systems (Greene Smith, 1993, 1994; Schaffer, 1984, 1985) the Financial Forecasting Using Genetic Algorithms 551 fitness function is decomposed into independent components, and different population elements are assigned to optimize each component. Fitness Function: Credit Assignment Most research on classification by GA ....

[Article contains additional citation context not shown here]

Greene, D. P., and S. F. Smith. 1993. Competition-ba sed induction of decision models from examples. Machine Learning 13(2/3):229257.

....does not provide any form of internal memory or chaining of rules, and because the rules all have a common right hand side, it is more appropriate to view our system as a genetic based machine learning system. We feel our work shares much in common with other such systems, most notably COGIN (Greene Smith, 1993) and GABIL (De Jong, Spears Gordon, 1993) The event prediction problem has been described in an earlier paper, but only a very brief description of the genetic algorithm was provided (Weiss Hirsh, 1998) In this paper we provide a detailed description of Timeweaver, our genetic based machine ....

....subset of the target events and collectively predict most of the target events. This is a Michigan style GA, since each individual pattern is only part of a complete solution. Other GA based systems have used this approach to learn disjunctive concepts from examples (Giordana, Saitta Zini, 1994; Greene Smith, 1993; McCallum Spackman, 1990) However, rather than simply forming a solution from all of the individuals in the population, we employ a second step. This step orders the patterns from best to worst, based primarily on the precision of their predictions, and prunes redundant patterns (i.e. those ....

[Article contains additional citation context not shown here]

Greene, D. P., and Smith, S. F. 1993. Competition-Based Induction of Decision Models from Examples. Machine Learning. 13: 229-257.

....in the Pittsburgh approach, a simple GA can be used as each single individual can already represent the whole multi modal concept. However, more complex genetic operations and chromosome representations have to be introduced. To alleviate these problems, some hybrid approaches have been proposed [10, 11, 12, 21]. However, a simpler strategy, which will be used in this paper, is to learn just one disjunct at a time. This separate and conquer approach has been commonly used in many classical (non GA based) rule learning algorithms [9] and was applied to the GA setting in SIA [24] It can effectively ....

....[9] and was applied to the GA setting in SIA [24] It can effectively restrict the size of the search space. Moreover, it obviates the need of speciation, which often involves various special techniques (such as the universal suffrage operator in REGAL [10] and the coverage based filter in COGIN [11]) Another advantage of SIA over many other methods is its ability to deal with continuous attributes. Systems like REGAL use a binary representation, which may become very long for continuous attributes and thus significantly slow down the GA process. On the contrary, SIA uses a high level ....

D.P. Greene and S.F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

....coming under the scrutiny of a substantial number of researchers. The task of constructing useful decision models (or rules or concepts) from examples (or instances, events) in realistic environments is complicated by problem complexity. Sources of problem complexity include the following (Greene, 1993): Generating function. Often the underlying behaviour comes from an unknown function that is nonlinear and that may have interacting, non homogeneous variables. Noise. Further complicating the model generation problem is the possibility of noise in the form of errors in recording the ....

....examples measured along a large number of dimensions with mixed data types. This gives rise to a very large space of possible models that must be searched. Many studies have tackled the problem of exponential searching. Some of them are CN2 (Clark and Niblett, 1989) GIL (Janikow, 1991) and COGIN (Greene, 1993). CN2 incorporates ideas from both Michalski s (1969) AQ and Quinlan s (1983) ID3 algorithms. It uses a similar framework to AQ11 and an information theoretic entropy measure to evaluate complex quality which prefers complexes covering a large number of examples of a single class and few examples ....

[Article contains additional citation context not shown here]

Greene, D.P.(1993). Competition-based induction of decision models from examples, Machine Learning, 13, 229-257.

....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 ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

....to encourage the GA to evolve a set of coadapted rules, rather than a single optima is a current hot topic in the GA community. While there are a number of approaches that might have been used, the one we ve chosen is based on the approach used by Greene and Smith in their COGIN system [44]. Some other approaches are described in Appendix A. The COGIN approach was selected because it explicitly evolves a population of rules that cover different niches, without requiring an explicit statement of the number of rules expected. We did not use the Pitt approach [60] in which each ....

....Disadvantage: Seems that the peaks must be the same height or they will not get picked up. As used by Smith et al. this approach uses a syntactic measure of fitness, but they claim that it wouldn t have to be. This approach to fitness sharing is illustrated in Figure A.5. COGIN Greene and Smith [44] developed another approach to encouraging niche formation. This method also uses two fitnesses for each string. The first, the raw fitness is calculated from the fitness function. The strings are ranked according to the raw fitness (from best to worst) and then the operative fitness is ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13(2/3):229--257, November/December 1993.

....by the goal of representing graphical classifiers. 1.4.1 COGIN There does not appear to have been a great deal of work on the application of genetic algorithms to classification learning. One effort in that direction, however, is Greene and Smith s COGIN (Coverage based Genetic Induction) [9]. COGIN uses a population of fixed length rules, where each rule gives a classification for some Boolean combination of attribute values. For an example rule: if (Attr 1 = val 6 val 7 ) Attr 2 = val 1 ) then class 3 The set of rules at any given generation comprises a classification model, ....

P.G. Greene and S.F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

....1993) Disjunctive Normal Form (DNF) concept descriptions are evolved using an LS 1 style approach. This work is aimed at a single class learning application. The goodness of a concept description is measured as the square of examples correctly classified. The COGIN approach developed elsewhere (Greene Smith 1993; 1994) addresses multi class problem domains introducing competition for coverage of training examples, encouraging the population to co operatively solve the concept learning task. Each rule is a conjunction of attribute value sets in binary coding. In this approach, the newly created rules ....

Greene, D. P., and Smith, S. F. 1993. Competition-- based induction of decision models from examples.

....theories. And finally in section 7 we discuss some further improvements to the methodology and conclude. 2 Background GA based systems in rule based concept learning may be divided into two main streams, based on their knowledge representation. The 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 ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 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 ....

....of multimodal concept learning has been dealt with in two different ways. In the Michigan approach each rule is encoded as a fix length individual, and the complete (multimodal) solution consists of several individuals in the population. Examples of systems following this approach are GOGIN [10], the Classifier systems [14] and the early versions of REGAL [7] In the Pittsburgh 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 ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

....promising substructures. GAs are, in general, successful in avoiding local minima and produce often near optimal solution, provided that they are given enough resources. Systems utilizing the search mechanism of GAs have recently been successfully applied to propositional concept learning, e.g. [15, 4, 10, 19]. However, only a few GA based systems are able to perform relational learning [10, 1, 16] and even they have been applied to relational domains sparingly. In this paper we shall describe the relational GA based system DOGMA [17, 18] that is capable of relational concept learning and theory ....

....a new fitness function that combines the minimal description length principle [31] and an information gain measure [27] 2 Background GA based systems in rule based concept learning may be divided into two main streams based on their knowledge representation. The Michigan type systems, e.g. [15], 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 deal with disjunctive concepts. In the Pittsburgh approach, e.g. 4, 19] the situation is reversed. These systems use chromosomes of ....

D. P. Greene and S. F. Smith. Competition-based induction of decision models from examples. Machine Learning, 13:229--257, 1993.

No context found.

D. P. Greene and S. F. Smith, "Competition-based induction of decision models from examples," Machine Learning, vol. 13, pp. 229--257, 1993.

No context found.

D.P. Greene, S.F. Smith, Competition-based induction of decision models from examples, Machine Learning 13 (1993) 229--257.

No context found.

D.P. Greene & S.F. Smith. Competition-based induction of decision models from examples. Machine Learning 13, 229-257. 1993.

No context found.

Greene, D. P. & Smith, S. F. Competition-based induction of decision models from examples. Machine Learning 13, 229-257 (1993).

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