Giordana, A., Saitta, L., and Zini, F. (1994) Learning disjunctive concepts by means of genetic algorithms. Proc. 11th International Conf. on Machine Learning, 96-104  Home/Search   Document Not in Database   Summary   Related Articles   Check

....noise by introducing some noise threshold for the score. Whenever the score of a formula falls below this threshold, the remaining uncovered examples are considered as noise. Finally, we have to emphasize that stochastic search in general has been used before in Machine Learning (see, e.g. [10, 16]) However, to the best of our knowledge, this is the first attempt to introduce algorithms from stochastic local search (SLS) into propositional Machine Learning. Reducing the problem of k term DNF to SAT, we can draw from a huge body of results from the area of satisfiability algorithms. In this ....

Giordana, A., Saitta, L., and Zini, F. (1994) Learning disjunctive concepts by means of genetic algorithms. Proc. 11th International Conf. on Machine Learning, 96-104

....noise by introducing some noise threshold for the score. Whenever the score of a formula falls below this threshold, the remaining uncovered examples are considered as noise. Finally, we have to emphasize that stochastic search in general has been used before in Machine Learning (see, e.g. [10, 16]) However, to the best of our knowledge, this is the first attempt to introduce algorithms from stochastic local search (SLS) into propositional Machine Learning. Reducing the problem of k term DNF to SAT, we can draw from a huge body of results from the area of satisfiability algorithms. In this ....

Giordana, A., Saitta, L., and Zini, F. (1994) Learning disjunctive concepts by means of genetic algorithms. Proc. 11th International Conf. on Machine Learning, 96-104 416

....evolving population. The idea of using genetic algorithms (GA s) or genetic programming for classification tasks in KDD is not new. In GABIL [2] a Genetic Algorithm is used so that the bit string represents a set of rules. Other approaches to data mining using evolutionary computing include REGAL [4] and GA MINER [3] The main difference between existing evolutionary approaches to data mining and the new algorithm presented in this paper is that we use a direct representation of solutions compared to an indirect representation. The decision trees are considered the building blocks of the ....

Giordana, A., Saitta, L., Zini, F. Learning Disjunctive Concepts by means of Genetic Algorithms. From Proceedings on the 11th International Conference on Machine Learning (ICML-

....for rule discovery. Although these genetic operators have been used mainly in the classification task, in general they can be also used in other tasks that involve rule discovery, such as dependence modeling. We review some of these operators in the following subsections. 3.2.1 Selection. REGAL [27] follows the Michigan approach, where each individual represents a single rule. Since the goal of the algorithm is to discover a set of (rather than just one) classification rules, it is necessary to avoid the convergence of the population to a single individual (rule) REGAL does that by using a ....

....rules, each of them covering a different part of the data space. 3.2.2 Generalizing Specializing Crossover. The basic idea of this special kind of crossover is to generalize or specialize a given rule, depending on whether it is currently overfitting or underfitting the data, respectively [27], 3] Overfitting was briefly discussed in sections 2.2.1 and 2.3.1. Underfitting is the dual situation, in which a rule is covering too many training examples, and so should be specialized. A more comprehensive discussion about overfitting and underfitting in rule induction (independent of ....

Giordana A, Saitta L, Zini F. Learning disjunctive concepts by means of genetic algorithms. Proc. 10th Int. Conf. Machine Learning (ML-94), 96-104. Morgan Kaufmann, 1994.

....of the authors [20] in that it handles positive and negative examples expressed as constrained clauses instead of definite clauses. This language of examples and hypotheses constitutes a major difference with most ILP learners, and in particular with FOIL[17] ML Smart [1] PROGOL [13] or REGAL [3]. Another major difference is that we aim at finding out the set Th(Ex) of all consistent clauses generalizing any given example Ex. In opposition, other learners construct concise theories, i.e. they retain the best clause in Th(Ex) in the sense of some numerical criterion (e.g. quantity of ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Cohen W. and Hirsh H., editors, Proceedings of ICML-94, pages 96--104. Morgan Kaufmann, 1994.

....do well at predicting a 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 ....

Giordana, A., Saitta, L., and Zini, F. 1994. Learning Disjunctive Concepts by Means of Genetic Algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, 96-104.

....value back to the rules along the activation chain when the system is doing well. The complex dynamics of this micro economy model results in emergent problem decomposition and the preservation of diversity. A more task specific approach to evolving a population of coadapted rules was taken by Giordana et al. 1994) in their REGAL system. REGAL learns classification rules consisting of conjunctive descriptions in first order logic from preclassified training examples. Problem decomposition is handled by a selection operator called universal suffrage, which clusters individuals based on their coverage of a ....

Giordana, A., Saitta, L. and Zini, F. (1994). Learning disjunctive concepts by means of genetic algorithms. In Cohen, W. and Hirsh, H., editors, Proceedings of the Eleventh International Conference on Machine Learning, pages 96--104, Morgan Kaufmann, San Francisco, California.

....consiste a ne remplacer un individu que par un individu meilleur. Une variante particuli ere consiste a distinguer en fonction du probl eme les individus sur lesquels doit agir la s election a chaque g en eration (les tories) des individus a conserver a plus longue ech eance (les whigs) [21]. M ethode Pratiquement, la s election est souvent effectu ee par tirage dit a la roulette. La roulette comprend P cases, la largeur de la case i etant proportionnelle a F (x i ) F . La boule est tir ee P fois ; lorsqu elle tombe dans la case i, une copie de l individu x i est ajout ee a ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Cohen W. and Hirsh H., editors, Proceedings of ICML-94, International Conference on Machine Learning, pages 96--104. Morgan Kaufmann, 1994.

....small resubstitution error. The overall approach would not work, if the clauses in the population were the same extensionally. In other words, there would be no division of labour among the clauses (C3) This is also the motivation for the universal suffrage selection algorithm presented in (Giordana et al. 1994). We took a simple extensiondriven approach to solve this problem: the algorithm only considers those refinements that yield clauses with an extension different from the extensions of clauses in the current population. The extension only has to differ in one example to make it different. This ....

....construction based on hypotheses returned by Progol (Muggleton 1995) For each clause, each input output connected subset of literals is used to define a feature. In contrast to all previously discussed methods, this method works for all types of background knowledge. In contrast to SP, REGAL (Giordana et al. 1994) is a concept learning algorithm. It is a full fledged genetic algorithm. REGAL s universal suffrage selection algorithm is the first extension driven approach to stochastic search in machine learning. MILP (Kovacic 1994) is an ILP algorithm that performs stochastic search for single clauses to ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, pages 96--104, 1994.

....data includes descriptions of hypothetical samples corresponding to 23 species of gilled mushrooms in the Agaricus and Lepiota family; each example is classified as edible or poisionous. The dataset was first used by Schlimmer [87] but has also been used by many other machine learning researchers [38, 39, 80, 84, 75, 110, 58] Soybean This dataset includes examples of soybean plants, each afflicted by one of 19 different diseases. The dataset was first used by Michalski and Chilausky [70] and has also been used frequently in the machine learning literature [69, 84, 97, 16, 102, 17] Congressional voting record ....

Attilio Giordana, Lorenza Saitta, and Floriano Zini. Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, New Brunswick, NJ, pages 96--97. Morgan Kaufmann, San Francisco, 1994.

....credit assignment) and that instead of forming a ruleset from the entire population, a second step is used to form a ruleset from a subset of the rules in the population. Our approach is also similar to the approach taken by other genetic algorithms which learn disjunctive concepts from examples (Giordana, Saita Zini 1994; McCallum Spackman 1990) We use a steady state GA instead of a generational GA because we expect the time to evaluate an individual to be large (due to large training sets) and a steady state GA is believed to be more computationally efficient in this case. The main difference between these ....

Giordana, A., Saitta, L., and Zini, F. 1994. Learning Disjunctive Concepts by Means of Genetic Algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, 96-104.

....The first steps in that directions on the problem of chromatography use at the moment a deterministic evolution criterion for the population of test cases, with promising preliminary results. ffl Other heuristics, proposed to tackle a similar difficulty in the framework of classifier systems [64], will be investigated: instead of accounting the different fitness cases solely by the mean square error on all fitness cases, the selection procedure can be modified in such a way that the best individuals on at least one fitness case remain in the population. ffl It has been experimentally ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Cohen W. and Hirsh H., editors, Proceedings of ICML-94, International Conference on Machine Learning, pages 96--104. Morgan Kaufmann, 1994.

.... via a simulated micro economy (Holland and Reitman 1978) Other extensions to current evolutionary paradigms have been proposed to encourage the emergence of niches and species in a single population (DeJong 1975; Deb and Goldberg 1989; Davidor 1991; Forrest, Javornik, Smith, and Perelson 1993; Giordana, Saitta, and Zini 1994; Moriarty and Miikkulainen 1996) in which individual niches compete for the allocation of trials. The use of multiple interacting subpopulations has also been explored as an alternate mechanism for coevolving niches using the so called island model (Grosso 1985; Cohoon, Hegde, Martin, and ....

Giordana, A., L. Saitta, and F. Zini (1994). Learning disjunctive concepts by means of genetic algorithms.

.... via a simulated micro economy (Holland and Reitman 1978) Other extensions to current evolutionary paradigms have been proposed to encourage the emergence of niches and species in a single population (De Jong 1975; Deb and Goldberg 1989; Davidor 1991; Forrest, Javornik, Smith, and Perelson 1993; Giordana, Saitta, and Zini 1994; Moriarty and Miikkulainen 1996) in which individual niches compete for the allocation of trials. The use of multiple interacting subpopulations has also been explored as an alternate mechanism for coevolving niches using the so called island model (Grosso 1985; Cohoon, Hegde, Martin, and ....

Giordana, A., L. Saitta, and F. Zini (1994). Learning disjunctive concepts by means of genetic algorithms.

....to the specialization operator. The overall approach would not work, if the clauses in the population were the same extensionally. In other words, there would be no division of labour among the clauses (R3) This is also the motivation for the universal suffrage selection algorithm presented in [6]. We took a simple extension driven approach to solve this problem: the algorithm only considers those refinements that yield clauses with an extension different from the extensions of clauses in the current population. This (extensional) restriction can also be used to enforce the construction ....

....node in the graph. The constructed features are either context dependent node attributes of depth n or context dependent edge attributes of depth n . This method also works in general (for graphs) but using fixed length paths obviously becomes prohibitive for large n. In contrast to SP, REGAL [6] is a concept learning algorithm. It is a fullfledged genetic algorithm. REGAL s universal suffrage selection algorithm is the first extension driven approach to stochastic search in machine learning. MILP [8] is an ILP algorithm that performs stochastic search for single clauses to overcome the ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, pages 96--104, 1994.

....the specialization operator. The overall approach would not work, if the clauses in the population were the same extensionally. In other words, there would be no division of labour among the clauses (R3) This is also the motivation for the universal suffrage selection algorithm presented in [ Giordana et al. 1994 ] We took a simple extension driven approach to solve this problem: the algorithm only considers those refinements that yield clauses with an extension different from the extensions of clauses in the current population. This (extensional) restriction can also be used to enforce the ....

....in the graph. The constructed features are either context dependent node attributes of depth n or context dependent edge attributes of depth n . This method also works in general (for graphs) but using fixed length paths obviously becomes prohibitive for large n. In contrast to SP, REGAL [ Giordana et al. 1994 ] is a concept learning algorithm. It is a full fledged genetic algorithm. REGAL s universal suffrage selection algorithm is the first extension driven approach to stochastic search in machine learning. MILP [ Kovacic, 1994 ] is an ILP algorithm that performs stochastic search for single clauses ....

A. Giordana, L. Saitta, and F. Zini. Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, pages 96--104, 1994.

....representation where examples may be described with a variable number of predicates. This paper introduces a new algorithm SIAO1 that is a generalization of SIA, a supervised learning system that uses attribute value representation only and not FOL (Venturini 1994) REGAL (Giordana Saitta, 1993) (Giordana, Saitta Zini, 1994) is the only known solution for learning rules in FOL with GAs. To deal with this representation obstacle, REGAL assumes that the user can provide a general model, called a template, of the formula to be learned, which we do not. In this paper, we shall have no place to compare with REGAL. Let us ....

....1991) For FOL representation, the problem is to find a good representation, i.e, a representation that would let the crossover play its role, and that could represent interesting rules for the user. REGAL is a successful attempt to deal with FOL representation in GAs (Giordana Saita 1993) (Giordana, Saitta Zini 1994). REGAL requires that the user provides a model of the rule to be learned. For instance, the user may provide the following model (Giordana Saitta 1993) Color(X, red,blue, Shape(X, square,triangle, 9 Y [color(Y, red,blue, far(X,Y, 0,1,2,3, Then, the GA is used to discover the ....

[Article contains additional citation context not shown here]

Giordana A., Saitta L., and Zini F. 1994. Learning disjunctive concepts by means of genetic algorithms, Proceedings of the 1994, Proceedings of the Eleventh International Conference on Machine Learning, 96-104.

.... approaches on many real world problems, such as the DNA promoter task (Cohen, 1992) There have been several genetic based, first order logic, multimodal concept learners (Greene Smith, 1993; Janikow, 1993) Giordana and Saitta (1993) showed how to integrate one of these system, Regal (Giordana, Saitta, Zini, 1994; Neri Saitta, 1996) with the deductive engine of ML SMART (Bergadano, Giordana, Ponsero, 1989) to help refine an incomplete or inconsistent domain theory. This version works by first using an automated theorem prover to recognize unresolved literals in a proof, then uses the GA based Regal ....

Giordana, A., Saitta, L., & Zini, F. (1994). Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the Eleventh International Conference on Machine Learning, pp. 96--104, New Brunswick, NJ. Morgan Kaufmann.

.... 1994 Floriano Zini worked for the Machine Learning and Genetic Algorithms Group at the Computer Science Department of the University of Torino (Italy) He participated to the design and implementation of REGAL, a system for learning multimodal concepts by means of distributed genetic algorithms [5, 6, 7, 8]. Since August 1995 to January 1996 he worked for the Logic Programming and Automated Reasoning Group at the Computer Science Department of the University of Torino (Italy) His research fucused on modal extension of logic programming languages. Since January 1996 to February 2001 he was a ....

A. Giordana, L. Saitta, and F. Zini. Learning Disjunctive Concepts by Means of Genetic Algorithms. In Proc. of 11th International Conference on Machine Learning, pages 96-104, New Brunswick, New Jersey, 1994. M. Kaufmann.

No context found.

A. Giordana, L. Saitta, and F. Zini. Learning Disjunctive Concepts by Means of Genetic Algorithms. In Proc. of 11th International Conference on Machine Learning, pages 96--104, New Brunswick, New Jersey, 1994. M. Kaufmann.

No context found.

Giordana, A., Saitta, L., and Zini, F. (1994) Learning disjunctive concepts by means of genetic algorithms. Proc. 11th International Conf. on Machine Learning, 96-104

No context found.

Attilio Giordana, Lorenza Saitta, and Floriano Zini. Learning disjunctive concepts by means of genetic algorithms. In Proceedings of the 11th International Conference on Machine Learning, pages 96-- 104, 1994.

No context found.

A. Giordana, L. Saitta & F. Zini (1994). Learning disjunctive concepts by means of genetic algorithms. Proc. 11th Intl. Conf. on Machine Learning, 96-104.

No context found.

A. Giordana, L. Saitta, F. Zini. Learning disjunctive concepts by means of genetic algorithms. Proc. 11 Int. Conf. Machine Learning (ICML-94), 96-104. Morgan Kaufmann, 1994.

No context found.

A. Giordana, L. Saitta, and F. Zini. Learning Disjunctive Concepts by Means of Genetic Algorithms. In Proc. Int. Conf. on Machine Learning, pages 96--104, 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