M. Sebag and C. Rouveirol. Tractable induction and classification in first order logic via stochastic matching. In Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI-97) , pages 888--893, San Francisco, CA, 1997. Morgan Kaufmann.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the search space can grow very quickly in ILP applications. Several techniques have therefore been proposed to improve search efficiency. Such techniques include improving computation times at individual nodes [4, 26] better representations of the search [3] sampling the search space [27, 28, 32], and parallelism [8, 13,19] Parallelism can be obtained from very different alternative approaches, such as dividing the search tree, dividing the examples, or even through performing cross validation in parallel [31] An intriguing alternative approach that can lead to better accuracy whilst ....

M. Sebag and C. Rouveirol. Tractable induction and classification in first-order logic via stochastic matching. In Proceedings of the 15th International Joint Conference on Artificial Intelligence, pages 888-893. Morgan Kaufmann, 1997.

....is not improvement in accuracy resulting from the use of B2 over B1 but the running time using B2 is lower than that for B1. Although an accuracy of 81 was achieved, the experiments carded out using Progol in (Srinivasan et al. 1996) show an average accuracy of 83 . Two recent approaches, (Sebag and Rouveirol, 1997) and (Kramer and De Raedt, 2001) have reported a maximum accuracy of 93.6 and 94.7 , respectively for mutagenesis database. The approach taken by Sebag and Rouveirol (1997) comes from the field of ILP and is concemed with polynomial induction and use of first order logic hypotheses with no size ....

....experiments carded out using Progol in (Srinivasan et al. 1996) show an average accuracy of 83 . Two recent approaches, Sebag and Rouveirol, 1997) and (Kramer and De Raedt, 2001) have reported a maximum accuracy of 93.6 and 94.7 , respectively for mutagenesis database. The approach taken by Sebag and Rouveirol (1997) comes from the field of ILP and is concemed with polynomial induction and use of first order logic hypotheses with no size restriction. Instead of exhaustively exploring the set of matchings between any example and any candidate hypothesis, the user determines the number of matchings samples to ....

Sebag, M. and Rouveirol, C. Tractable Induction and Classification in First Order Logic via Stochastic Matching. In Proceeding of the 15 th International Joint Conference on Artificial Intelligence, 1997.

....in Prolog by asserting R and running the query class(e, class) thus X being instantJared with the example key) This boils down to meta calling the query once( 1, In) with the variable X instantiated. k local) Horn clauses. Other papers can be found in the area of stochastic evaluation [Sebag and Rouveirol, 1997; Srinivasan, 1999] The idea there is that one requires a hypothesis to be true (or false) with a high probability through the use of stochastic sampling and matching, which allows for significant speedups but changes the results. More recently, some new (exact) optimizations have been proposed. ....

M. Sebag and C. Rouveirol. Tractable Induction and Classification in First-Order Logic via Stochastic Matching. In Proceedings of the 15th International Joint Conference on Artificial Intelligence, pages 888-893. Morgan Kaufmann, 1997.

....sets or decision trees. Also, the generated models cannot be reformulated into first order theories; these learners can therefor not be used to solve relational learning problems with multipleinstance propositionalized data. Because propositionalization based relational learners (such as STILL [14]) often outperform classical relational learners, relational learning based on multiple instance propositionalization (which is much more adapted to nondeterminate domains than standard propositionalization [17] is a promising field for which efficient multiple instance rule learners will be ....

Michele Sebag and Celine Rouveirol. Tractable induction and classification in first order logic. In IJCAI, Nagoya, Japan, 1997.

....function free program clauses. In JGA and DOGMA the hypothesis space is restricted through the definition of a language template that restricts the form of possible hypothesis considered. A different approch to the matching problem of first order logic is to avoid it by using stochastic matching [109, 45]. In this approach one does not exhaustively consider all possible matchings between variables and objects, but instead one takes a stochastic sample of the matchings. Therefore, one can not always definitely say whether a formula OE covers an example e. Instead the matching is probabilistic and ....

M. Sebag and C. Rouveirol. Tractable induction and classification in first order logic via stochastic matching. In Proceedings of the 15th International Joint Conference on Artifical Intelligence, pages 888-- 892. Morgan Kaufmann, 1997.

....reasons that motivate us for finding such algorithms are that MMPs play a central role in learning structure activity relations. This is the problem that was solved in the REMO learning system (Zucker and Ganascia 1994; Zucker and Ganascia 1996) REPART (Zucker, Ganascia et al. 1998) and STILL (Sebag and Rouveirol 1997) Inductive Logic Programming systems. Section 2 is a more formal presentation of the MIP problem, shows how it is linked to the MPP problem and explains how in the two cases problem solving comes down to learning special concepts called multiple ones. Section 3 proposes extensions to classical ....

.... the heart of the REMO system which enables the efficient learning of relations from several thousand structured examples (Zucker and Ganascia 1996) In order to build a disjunctive version space, the STILL system solves an MPP problem iteratively, in which it takes one positive example at a time (Sebag and Rouveirol 1997). This system has obtained the best results for the ILP problem of mutagenesis (Srinivasan, Muggleton et al. 1997) In MPP, as in MIP, each example is represented by a bag of instances. In MIP, an instance is a snapshot of the entire object, whereas in MPP, an instance is a small part of the ....

Sebag, M. and C. Rouveirol 1997. Tractable Induction and Classification in First Order Logic. Fifteenth International Joint Conference on Artificial Intelligence, IJCAI'97, Nagoya, Japan, Morgan Kaufmann.

....we perform a simple form of predicate invention through first order feature construction, and use the constructed features for propositional learning. The line of research related to LINUS is reported by Zucker and Ganascia [56] Fensel et al. 11] Geibel and Wysotzki [18] Sebag and Rouveirol [48] and others. Other related approaches to the propositionalisation of relational learning problems include Turney s RL ICet al..gorithm [52] Kramer et al. s [24] stochastic predicate invention approach and Srinivasan and King s [50] approach to predicate invention achieved by using a variety of ....

M. Sebag and C. Rouveirol. Tractable induction and classification in firstorder logic via stochastic matching. In Proceedings of the 15th International Joint Conference on Artificial Intelligence, pages 888--893. Morgan Kaufmann, 1997.

....(a) reducing the time for the AND OR proof for the examples. These can be achieved by either storing the results of earlier proofs (as is done in [5] or by recoding the background knowledge to have a lower branching factor and depth, or the use of stochastic matching techniques (as is done in [34]) and (b) reducing the number of examples, either by not including some kinds of examples (for example, positive only learning as described in [25] or by sampling small fractions of the examples which is the remit of this paper. For illustrative reasons, Fig. 3 tabulates the orders of ....

M. Sebag and C. Rouveirol. Tractable Induction and Classification in First-Order Logic via Stochastic Matching. In Proceedings of the Fifteenth International Conference on Artificial Intelligence (IJCAI-97). Morgan Kaufmann, Los Angeles, CA, 1997.

.... FOCL system using relational clich is an ad hoc solution to avoid the problem of local minima of information gain [16] Different promising approaches also propose a new stochastic bias, either on the exploration of the search space of the concept, SFOIL[15] or on the subsumption itself, STILL[17]. The genetic algorithm based REGAL system reports also good results [18] To our knowledge, PROGOL is one of the most efficient system learning nondeterminate clauses. PROGOL allows the programmer to characterize very precisely the degree of indeterminacy using lists of modes [19] Nevertheless, ....

Sebag, M. and C. Rouveirol. Tractable Induction and Classification in First Order Logic. in Fifteenth International Joint Conference on Artificial Intelligence, IJCAI'97. 1997. Nagoya, Japan: Morgan Kaufmann.

....employs the more conventional separate and conquer strategy. In principle, SP is capable of exploring features of arbitrary length, but in practice the constraints are strong enough to keep search focused. The only other algorithm capable of efficiently exploring arbitrarylength features is STILL (Sebag and Rouveirol 1997). Interestingly, STILL is also a stochastic algorithm, but stochastic search is applied in a totally different manner in STILL, namely in the matching phase between features and examples. STILL also uses a totally different representation, which effectively yields black box classifiers instead of ....

M. Sebag and C. Rouveirol. Tractable induction and classification in first order logic via stochastic matching. In Proc. Fifteenth International Joint Conference on Artificial Intelligence (IJCAI-97), pages 888--893, San Mateo, CA, 1997. Morgan Kaufmann.

....(1995) The crucial point is how to efficiently evaluate that no progress can be made 17. In representation languages that extend flat feature vectors, such as first order logic, the complexity of a learning problem also depends crucially on the average cost of matching an instance with a rule. Sebag and Rouveirol (1997) demonstrate a technique for reducing these potentially exponential costs via sub sampling. F urnkranz by shifting to a more complex hypothesis space. In the propositional case, forward selection approaches to feature subset selection, i.e. algorithms that select the best subset of features by ....

Sebag, M., & Rouveirol, C. (1997). Tractable induction and classification in first order logic via stochastic matching. In Proceedings of the 15th International Joint Conference on Artificial Intelligence (IJCAI'97), pp. 888--893, Nagoya, Japan.

.... (ILP) directly in a FOL framework by adopting a generate and test approach carefully controlled through sophisticated language and search bias (see for instance, ij determination [11] see [12] for an extended study of bias in ILP) A number of ILP systems (among others, LINUS [8] STILL [17], REPART [21] SP [7] have adopted an alternative to this approach and have addressed the problem by first reformulating the initial FOL learning problem into an attribute value or boolean problem and then by applying efficient learning techniques dedicated to this simpler formalism. This ....

....may be successfully applied, provided that the discriminant features of the FOL learning problem are preserved by propositionalisation. But, as the subsumption test can be exponential even in restricted of FOL languages such as Datalog, the reformulated training set can be of exponential size [17], as well as highly 1 Inference and Learning Group, Laboratoire de Recherche en Informatique, UMR 8623 of CNRS, Batiment 490, Universite Paris Sud 91405 Orsay Cedex (France) email : falphonse,celineg lri.fr redundant, and cannot be directly addressed as such for complex relational learning ....

[Article contains additional citation context not shown here]

M. Sebag and C. Rouveirol, `Tractable induction and classification in first order logic via stochastic matching', in 15th Int. Join Conf. on Artificial Intelligence (IJCAI'97), Nagoya, Japon, pp. 888--893. Morgan Kaufmann, (1997).

No context found.

M. Sebag and C. Rouveirol. Tractable induction and classification in first order logic via stochastic matching. In Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI-97) , pages 888--893, San Francisco, CA, 1997. Morgan Kaufmann.

No context found.

Sebag, M., Rouveirol. C.: Tractable Induction and Classification in First Order Logic via Stochastic Matching. In: Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (IJCAI-97). Morgan Kaufmann, San Francisco, CA (1997) 888-893

No context found.

Sebag, M., & Rouveirol, C. (1997). Tractable induction and classification in first order logic via stochastic matching. Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (pp. 888-- 893). Nagoya, Japan: Morgan Kaufmann.

No context found.

Sebag, M., & Rouveirol, C. (1997). Tractable induction and classification in first order logic via stochastic matching. Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (pp. 888-- 893). Nagoya, Japan: Morgan Kaufmann.

No context found.

Sebag, M., & Rouveirol, C. (1997). Tractable induction and classification in first order logic via stochastic matching. Proceedings of the Fifteenth International Joint Conference on Artificial Intelligence (pp. 888-- 893). Nagoya, Japan: Morgan Kaufmann.

No context found.

Automatica, 14:465--471. Sebag, M. (1997). Tractable induction and classification in first order logic. In 15th International Joint Conference on Artificial Intelligence, Tokyo, Japan.

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