Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, Catania, Italy, 1994. SpringerVerlag.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... or [Esposito et al. 1993] Pre pruning heuristically deciding when to stop growing clauses and concepts has been present in Inductive Logic Programming (ILP) in the form of stopping criteria for quite some time (see e.g. Foil [Quinlan, 1990] mFoil [Dzeroski and Bratko, 1992] and Fossil [Furnkranz, 1994]) The basic idea behind most post pruning methods is to learn a concept description on one part of the training instances and to subsequently delete several parts of this theory in order to improve performance on the remaining set. The most prominent use of this method in ILP is the adaptation of ....

....pruning these clauses. We solve this problem by adapting the relational learning algorithm Fossil to combine pre pruning and post pruning by first performing a general to specific search for a good starting theory and then pruning this theory. 2 Fossil and the Cutoff Stopping Criterion Fossil [Furnkranz, 1994] is a Foil like ILP system that uses a search heuristic based on statistical correlation. Intuitively, the so called correlation coefficient corr(c; l) measures the congruence of the truth values of the instances covered by the partially grown clause c with the truth values assigned to these ....

[Article contains additional citation context not shown here]

Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, Catania, Italy, 1994. SpringerVerlag.

....a rule and the overall distribution of positive and negative examples by comparing the likelihood ratio statistic to a 2 distribution with 1 degree of freedom at the desired significance level. Insignificant rules are rejected. ffl Cutoff Stopping Criterion: This simple method used in Fossil [Furnkranz, 1994a] only adds a condition to a rule when its heuristic value is above a predefined threshold. mFoil s significance testing along with the m estimate and a powerful beam search have been very successful in learning concepts in noisy domains [Dzeroski and Bratko, 1992] Similar results have been ....

....for the very efficient cutoff criterion. Both have been shown to be superior to the encoding length restriction, because the latter is dependent on the size of the training set, so that the size of the learned concepts (and thus the amount of overfitting) may increase with training set size [Furnkranz, 1994a] 3.2 Post Pruning Post pruning was introduced to relational learning algorithms with Reduced Error Pruning (REP) Brunk and Pazzani, 1991] based on previous work by [Quinlan, 1987] and [Pagallo and Haussler, 1990] The basic idea is that in a first pass, no attention is payed to the noise in ....

[Article contains additional citation context not shown here]

Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, pages 122--137, Catania, Italy, 1994. Springer-Verlag.

....for achieving high accuracy can be given. 1 Introduction Lately several pruning methods for noise handling in relational rule learning algorithms have been proposed. The classic approaches to pruning are based on pre pruning (Foil [Quinlan, 1990] mFoil [Dzeroski and Bratko, 1992] or Fossil [Furnkranz, 1994b] and post pruning (Reduced Error Pruning (REP) Brunk and Pazzani, 1991] and Grow [Cohen, 1993] More recently approaches have been proposed that combine (MDL Grow [Cohen, 1993] and Top Down Pruning (TDP) Furnkranz, 1994c] and integrate (Incremental Reduced Error Pruning (I REP) Furnkranz ....

....and the overall distribution of positive and negative examples by comparing the likelihood ratio statistic to a 2 distribution with 1 degree of freedom at the desired significance level. Insignificant rules are rejected. ffl Cutoff Stopping Criterion: This simple method used in Fossil [Furnkranz, 1994b] only allows to add conditions to a rule when their heuristic values are above a predefined threshold. Thus this cutoff parameter allows the user to directly control the amount of overfitting. At Cutoff = 0:0 the algorithm will fit all of the data (no pre pruning) while at Cutoff = 1:0 Fossil ....

[Article contains additional citation context not shown here]

Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, pages 122--137, Catania, Italy, 1994. Springer-Verlag.

....of classification and regression in relational learning. 1 Introduction Learning to predict discrete classes from preclassified examples has long been, and continues to be, a central research topic in Inductive Logic Programming (e.g. Quinlan, 1990, Pazzani Kibler, 1992, Muggleton, 1995, Furnkranz, 1994]) Recently, there has also been increased interest in relational regression, i.e. the task of learning to predict continuous numeric variables from relational data (e.g. Karalic, 1995,Karalic Bratko, 1997,Kramer, 1996] A class of problems between classification and regression is learning ....

Furnkranz, J. (1994). Fossil: A robust relational learner". In Proceedings of the 7th European Conference on Machine Learning (ECML-94). Berlin: Springer Verlag.

.... e.g. 13] or [7] Two basic approaches can be discerned [1] Pre pruning heuristically deciding when to stop growing clauses and concepts has been present in Inductive Logic Programming (ILP) in the form of stopping criteria for quite some time (see e.g. Foil [18] mFoil [6] and Fossil [10]) The basic idea behind most post pruning methods is to learn a concept description on one part of the training instances and to subsequently delete several parts of this theory in order to improve performance on the remaining set. The most prominent use of this method in ILP is the adaptation ....

....show the applicability of this algorithm (Section 4) and from their results we draw some conclusions (Section 5) 1 Austrian Research Institute for Artificial Intelligence, Schottengasse 3, A 1010 Vienna, Austria. E mail: juffi ai.univie.ac. at 2 Fossil AND THE CUTOFF STOPPING CRITERION Fossil [10] is a Foil like ILP system that uses a search heuristic based on statistical correlation. Intuitively, the so called correlation coefficient corr(c; l) measures the congruence of the truth values of the instances covered by the partially grown clause c with the truth values assigned to these ....

[Article contains additional citation context not shown here]

Johannes Furnkranz, `Fossil: A robust relational learner', in Proceedings of the European Conference on Machine Learning, Catania, Italy, (1994). Springer-Verlag.

.... algorithm of Figure 1 is that we get a series of different concept descriptions in a roughly general to specific order (top down) 3 as opposed to pruning methods that generate a most specific theory first and then 2 A detailed description of the domain can be found in the Appendix of [9]. 3 Note that we use the terms general and specific in an intuitive way. We consider the empty theory to be most general, because Everything is false. is a very general statement. However, our most specific theory will cover more ground instances than the empty theory, and thus may be ....

Johannes Furnkranz, `Fossil: A robust relational learner', Technical Report OEFAI-TR-93-28, Austrian Research Institute for Artificial Intelligence, (1993). Extended version.

....also in terms of several other accuracy measures. Furthermore, we performed experiments in the domain of finite element mesh design (for details see [6] where the background knowledge is non determinate. Table 5 shows the results of SRT for the mesh dataset together with the results of FOSSIL [11] and results of other methods that were directly taken from [15] SRT performs better than FOIL [20] and mFOIL [8] but worse than the other methods. However, statistical analysis shows that only the differences between FOIL and the other algorithms are significant. 5 Struct. FOIL mFOIL GOLEM ....

J. Furnkranz, `Fossil: A robust relational learner', in Machine Learning: ECML-94, eds., F. Bergadano and L. De Raedt, pp. 122--137, Berlin Heidelberg New York, (1994). Springer.

.... subsequently be generalized by cutting off branches of the decision tree (as in [Quinlan, 1987] or [Breiman et al. 1984] In Inductive Logic Programming, pre pruning has been common in the form of stopping criteria as used in FOIL [Quinlan, 1990] mFOIL [D zeroski and Bratko, 1992] or FOSSIL [F urnkranz, 1994a] Post pruning was introduced to ILP with Reduced Error Pruning (REP) Brunk and Pazzani, 1991] based on ideas by [Quinlan, 1987] and [Pagallo and Haussler, 1990] First the training set is split into two subsets: a growing set and a pruning set. A concept description explaining all of the ....

Johannes F urnkranz. FOSSIL: A robust relational learner. In Proceedings of the European Conference on Machine Learning, pages 122--137, Catania, Italy, 1994. Springer-Verlag.

....that this threshold is robust in the sense that a good value for this parameter is independent of the number of training examples and of the amount of noise in the data (section 3. 3) Parts of this chapter have been previously published in a somewhat different form in [Furnkranz, 1993a] and [Furnkranz, 1994b] 3.1 Pre Pruning in Relational Learning As we have discussed in section 2.3.4, noise in the data is a problem for the simple SeparateAndConquer rule learning algorithm of figure 2.3, because it tries to find explanations for every single example in the training set, including the erroneous ....

Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, pages 122--137, Catania, Italy, 1994. Springer-Verlag.

.... This theory will subsequently be generalized by cutting off branches of the decision tree (as in [Quinlan, 1987] or [Breiman et al. 1984] In ILP, pre pruning has been common in the form of stopping criteria as used in Foil [Quinlan, 1990] mFoil [Dzeroski and Bratko, 1992] or Fossil [Furnkranz, 1994a] Post pruning was introduced to ILP with Reduced Error Pruning (REP) Brunk and Pazzani, 1991] based on ideas by [Quinlan, 1987] and [Pagallo and Haussler, 1990] First the training set is split into two subsets: a growing set and a pruning set . A concept description explaining all of the ....

Johannes Furnkranz. Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, Catania, Italy, 1994. Springer-Verlag.

No context found.

Furnkranz, J. (1994b). Fossil: A robust relational learner. In Proceedings of the European Conference on Machine Learning, Catania, Italy, pp. 122--137. Springer-Verlag.

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